The impact of vertical integration on health care delivery and costs: Evidence from physician–pharmacy integration Authors: Pragya Kakani Published: 2025 ABSTRACT AbstractVertical integration among health care providers is an increasingly common feature of U.S. health care. This study investigates the impact of vertical integration in the context of oncology practices launching in-house pharmacies dispensing high-cost oral cancer treatments using a stacked event study design. I find physician–pharmacy integration lowers point-of-sale drug prices paid by plans due to 1.0% lower prices at in-house pharmacies. I also find physician–pharmacy integration increases the number of patients filling new prescriptions by 6.2% and reduces time-to-fill for new prescriptions requiring prior authorization. These effects may arise because in-house pharmacies individually have limited bargaining power and because vertical integration decreases coordination costs, reflected by faster prior authorization.Keywords: Vertical integration, Prescription drugs, In-house pharmacies, Medically-integrated dispensing, Specialty pharmacies, Oncology, Medicare Part D, Prior authorization, 340B Drug Discount ProgramJEL classification: I10, I11, L22, L23 Introduction In recent years, there has been considerable growth in vertical integration among health care providers in the US, leading to the rise of conglomerate health systems encompassing hospitals, physician groups, post-acute care facilities, and other providers (Nikpay et al., 2018; Cutler et al., 2020; Beaulieu et al., 2023). Nonetheless, the effects of this phenomenon are poorly understood and theoretically uncertain. On the one hand, vertical integration can improve care quality by aligning financial incentives or reducing costs of coordination (David et al., 2013; Carlin et al., 2015; Rahman et al., 2016). On the other hand, aligning financial incentives may cause providers to promote low-value care (Afendulis and Kessler, 2007; Baker et al., 2016). There is also significant concern that vertical integration may cause anti-competitive price increases,1 or foreclose competition.2 While prior work has focused primarily on hospital acquisition of physician groups and post-acute care facilities, this study uses novel data to evaluate an increasingly prevalent but understudied form of vertical integration in the US: between physician organizations and pharmacies. This phenomenon is important for several reasons. First, physician organizations are increasingly launching and using in-house pharmacies to dispense high-cost, self-administered drugs. Between 2006 to 2019, I find the share of oral oncology spending filled by in-house pharmacies increased from 10% to 37% in the Medicare Prescription Drug (Part D) Benefit, consistent with prior work (Kanter et al., 2022; Kakani et al., 2024). Second, as “ultra-expensive” drugs have proliferated in recent years there are significant concerns regarding cost, access, and adherence to these therapies that may be impacted by physician–pharmacy integration (Kang et al., 2021; Dusetzina et al., 2022). Indeed, prior evidence on infused drugs suggests profitability can impact treatment choices (Jacobson et al., 2010; Bond et al., 2023; Horn, 2025) and physician drug dispensing in Switzerland has been found to be associated with higher drug spending in certain studies (Kaiser and Schmid, 2016; Trottmann et al., 2016; Burkhard et al., 2019). Partly due to these concerns, certain states restrict the operation of such pharmacies (Munger et al., 2014), and Medicare and Medicaid prohibited certain physician groups from operating mail-order pharmacies in 2023 (Mullangi, 2024). These restrictions are controversial, as many claim integration improves access and quality of care. As one physician notes: “Prior to our in-house pharmacy, we had no idea what happened after we sent prescriptions…Did the patient start treatment later than recommended? Did the patient take the right combination and on a consistent basis? These were all questions that affect quality and outcomes that we were not able to answer before. Now we can”. – Dr. Kerin Adelson, Chief Quality Officer, Yale Cancer Center (The ASCO Post, 2017) This study leverages new data on in-house pharmacies and Medicare Fee-for-Service (FFS) and Part D claims to evaluate the effects of integration between physician organizations and pharmacies on outcomes related to prices and patient care. Specifically, I use a stacked event study evaluating outcomes at 204 medical oncology practices clinically integrating with in-house pharmacies from 2009–2016 versus control practices. I focus on outcomes related to high-cost oral cancer treatments as in-house pharmacies specialize these medicines (Kakani et al., 2024). This analysis reveals that despite vertical integration being associated with price increases in other health care markets, physician–pharmacy integration reduces average point-of-sale prices paid to pharmacies. This is because Medicare Part D plans negotiated 1.0% lower point-of-sale prices with in-house pharmacies for high-cost oral cancer treatments, including with in-house pharmacies linked to the largest practices and health systems. These effects are similar among plans managing both medical and pharmacy benefits, illustrating that joint bargaining for pharmacy and medical services is not sufficient to increase pharmacy prices. This deviation from other contexts may be because even the largest in-house pharmacies individually have limited bargaining power with health insurers and pharmacy benefit managers (PBMs), who negotiate prices and design pharmacy networks on behalf of health plans. PBMs are themselves also vertically integrated with specialty pharmacies and thus have an additional incentive to suppress pricing at in-house pharmacies to deter in-house pharmacies from joining their networks and in-turn steer patients to their own pharmacies. Indeed, government reports suggest the large PBMs pay external pharmacies less than their own pharmacies for certain specialty medications and often offer unaffiliated pharmacies challenging contract terms (FTC, 2024). Physician–pharmacy integration also has important effects on utilization and quality of care during treatment initiation, which are likely due in part to operational efficiencies achieved via integration. Specifically, integration increases the likelihood of patients having a first-time fill for a high-cost oral cancer treatment by 6.2%. However, integration does not cause detectable substitution to higher cost regimens or away from low-cost oral chemotherapies or infused chemotherapies. Similarly, integration reduces the time-to-fill, or the time between the patient’s first fill of a drug and their last office visit, during which the prescription was likely written, with patients being 21.8% more likely to have a first fill within 5 days of their last office visit. However, integration does not improve medication adherence after treatment initiation. In probing the mechanisms underpinning effects, I find evidence of operational efficiencies associated with integration, namely faster prior authorization; integration reduces time-to-fill only when prior authorization is required. The key identification assumption is that treated and control outcomes would have trended in parallel without integration. Several careful methodological choices and robustness tests are used to develop confidence in this assumption. Placebo tests show that integration events do not coincide with other apparent changes in practice organization (e.g., practice size, number of patients treated) or patient selection (e.g., patient diagnoses, demographics). While unobservable changes coinciding with integration events cannot be ruled out, these are unlikely to mediate effects given the insensitivity of results to excluding observable covariates (Oster, 2019). The control group is also validated as a reasonable counterfactual in several ways. In addition to observing parallel pretrends, the control group does not include practices integrating at an earlier time, as this can contaminate estimated treatment effects (Goodman-Bacon, 2021; Callaway and Sant’Anna, 2021). Results are also insensitive to the choice of control group. Finally, the stacked event study design addresses identification issues arising when treatment timing varies (Sun and Abraham, 2021). These results suggest physician–pharmacy integration increases and speeds treatment initiation, partly due to operational efficiencies, and reduces point-of-sale drug prices. While this is among the first evaluations of physician–pharmacy integration in the US, these results differ in part from limited prior evidence. The most comparable, contemporaneous study, namely Kanter et al. (2024), evaluated the impact of independent oncology practices integrating with pharmacies on the use of oral cancer therapies. Using data from one commercial insurer in fourteen states, they found no overall impact on the likelihood of oral chemotherapy use, medication adherence, costs, or the time between diagnosis and treatment initiation. This study builds on this prior work by studying a larger, national sample of patients and practices, including independent and hospital-affiliated practices, by studying Medicare, where pharmacy network regulations may somewhat reduce plans’ ability to exclude in-house pharmacies (FTC, 2024; Kakani et al., 2025), by focusing on utilization of high-cost oral cancer treatments, which are far more likely to be supported by in-house pharmacies, and by considering certain additional outcomes that may be directly impacted (e.g., the time between prescription writing and treatment initiation and point-of-sale prices). These differences in sample and outcomes may explain differences in certain findings. These findings make several important contributions. First, this study is the largest evaluation of the impacts of physician–pharmacy integration, which is an increasingly important form of integration for high-cost medicines (Kanter et al., 2022; Kakani et al., 2024). Thus, this paper contributes to an important literature studying the impact of “buy-and-bill” payment models for medicines (Jacobson et al., 2010; Bond et al., 2023; Horn, 2025), drivers of medication non-adherence for specialty medications (Dusetzina et al., 2022), and the determinants of pharmacy prices (Lakdawalla and Yin, 2015; Starc and Swanson, 2021). Second, while prior work suggests price-increasing effects of vertical integration, these results show that vertical integration can yield lower prices. This may be particularly likely in cases where vertical integration reflects market entry of a competitor with limited bargaining power. Finally, while prior work has largely failed to find operational efficiencies from vertical integration (Post et al., 2018; Machta et al., 2019; Amado et al., 2022), this paper shows that vertical integration can achieve such efficiencies. These may be especially achievable during complex care transitions between otherwise separate entities, such as initiating a high-cost oral cancer treatment. Nonetheless, the aggregate impact on welfare remains uncertain and will depend critically on the value of marginal prescriptions. A review of 25 cost-effectiveness studies for high-cost oral cancer treatments, performed herein, reveals a wide range of estimates for cost-effectiveness that vary by drug, unobserved clinical factors (e.g., stage and biomarkers), and time period. The median weighted average incremental cost per quality-adjusted life year (QALY) is $167K (IQR: $90K to $266K), but the welfare effects vary depending on the assumed social willingness-to-pay for a QALY, over which there is significant controversy (Hirth et al., 2000; Braithwaite et al., 2008; Neumann et al., 2014). Finally, uncertainty remains because it is unclear how valuable utilization is for marginal patients, and this is an important area for future research. The remainder of the paper proceeds as follows: Section 2 reviews institutional context, Section 3 details a framework for evaluating the effects of physician–pharmacy integration, Section 4 reviews data, Section 5 reviews study design, Section 6 reviews outcomes, Section 7 reviews price effects, Section 8 reviews patient care effects, Section 9 discusses welfare, and Section 10 concludes. Institutional background Prescription drug supply chain The U.S. prescription drug supply chain is complex as illustrated in Appendix Section A, Appendix Figure 1. Pharmacies generally negotiate drug acquisition costs with wholesalers and point-of-sale prices with health plans. Health plans and pharmacies often negotiate through contracted intermediaries to improve bargaining power; plans contract with PBMs, and smaller pharmacies contract with pharmacy services administrative organizations. Point-of-sale prices represent the payment received by the pharmacy upon dispensing a specific drug, paid in part by patients and in part by the plan. Pharmacies can also be charged direct and indirect remuneration (DIR) payments by the plan after the sale, which reflect performance penalties (or bonuses) and other adjustments.3 The profit associated with the marginal prescription is thus the point-of-sale price – acquisition costs – pharmacy DIR fees. Prior work using public data on retail pharmacies’ acquisition costs suggests retail pharmacies receive 3%–4% average profit margins on branded drugs (Sood et al., 2017).4 Profits for in-pharmacies that are owned by hospitals enrolled in the 340B Drug Pricing Program are likely substantially higher as these entities acquire drugs at 20%–50% discounts (Desai and McWilliams, 2018). Over the last 10–20 years there has been increasing horizontal integration across PBMs as well as vertical integration between PBMs, health plans, and pharmacies. Four PBMs (Optum, CVS, Express Scripts-Cigna, Humana Pharmacy Solutions5) captured over 90% of the PBM market in Medicare nationally in 2023 (Qato et al., 2024; MedPAC, 2025). Each of these firms also operates vertically integrated health plans and pharmacies. PBM-owned pharmacies capture an especially large share of the high-cost, specialty drug market and often deliver these medications by mail. In Medicare, PBM-owned pharmacies captured nearly 40% of the specialty pharmacy market in 2021 (Kakani et al., 2025). PBM-owned pharmacies are partly able to maintain high market share in the specialty drug market because PBMs steer patients to their own pharmacies. This can be done by excluding unaffiliated pharmacies from networks, offering unaffiliated pharmacies low reimbursement to discourage network participation, placing affiliated pharmacies in preferred networks, and direct patient outreach (FTC, 2024). Medicare part D Medicare Part D covers the vast majority of oral cancer treatments for Medicare beneficiaries.6 This stands in contrast to Medicare Part B, which covers infused cancer treatments.7 To obtain Part D coverage, patients can select a stand-alone Part D plan, which only covers prescription drugs, and pursue medical coverage either through the Medicare Fee-for-Service or a separate Medicare Advantage Plan. Alternatively, patients may opt for a single Medicare Advantage-Prescription Drug (MA-PD) plan that covers medical services and prescription drugs. Medicare Part D has several unique features relative to other payers, relevant to studying physician–pharmacy integration in oncology. First, pharmacy networks are governed by an “Any Willing Pharmacy” requirement, which requires that plans cover all pharmacies that can satisfy standardized terms and conditions set by the plan. Thus, Medicare Part D plans are likely less able to exclude in-house pharmacies from their networks than most commercial plans (FTC, 2024). However, Medicare Part D plans may still retain some ability to effectively exclude pharmacies from networks by offering the lowest allowable reimbursement or creating stringent requirements for inclusion (e.g., licensing requirements, requirements that pharmacies carry certain products or have certain levels of on-site inventory, etc.) (National Association of Specialty Pharmacy, 2017). Second, Medicare Part D plans cannot exclude any oncology medicines from coverage but are permitted to require prior authorization whereby physicians must submit documentation to the plan demonstrating that the patient requires the drug before the drug will be covered. This may include evidence of an appropriate diagnosis and compliance with other requirements, such as trying other therapies first. Third, the Part D benefit has historically been divided into four phases of eligibility that have bearing on plan, patient, manufacturer and government liability. The bulk of spending on high-cost oral oncology drugs occurs in the “catastrophic phase”, in which the government, plan and patient liabilities were 80%, 15%, and 5% respectively over the study period (Dusetzina and Keating, 2016).8 Fourth, cost-sharing obligations cannot be paid through coupons often provided by manufacturers. Nonetheless, patients can apply for patient assistance programs from charitable organizations, often funded by manufacturers, to support cost-sharing (Dafny et al., 2022). Finally, some patients are eligible for the low-income subsidy program, through which premiums and cost-sharing are heavily subsidized. Regulation of in-house pharmacies Physician organizations generally have broad authority to procure and dispense drugs for their patients either via in-office dispensing or launching a state-licensed pharmacy. Staff and capability requirements for state-licensed pharmacies are usually stricter, but licensure can come with several benefits. Licensure of in-house pharmacies may be required by some states or payers to fill prescriptions from external practices, to deliver medicines by mail, or for plan network inclusion. The federal Anti-Kickback Statute and the prohibition against self-referral, commonly known as the Stark Law, generally permit in-office dispensing and physician-owned pharmacies, as both regulations contain exceptions for in-office ancillary services. Furthermore, no states currently have restrictions on health systems’ ability to own licensed pharmacies. Nonetheless, some states and federal agencies impose restrictions on independent, physician-owned practices ability to procure and dispense medications. These include restrictions on physician dispensing, physician-ownership of licensed pharmacies, or both. For example, dispensing is only permitted in small quantities or for special circumstances in Texas, New York, New Jersey, Montana, and Massachusetts (Munger et al., 2014). Certain states also restrict pharmacy ownership by physicians. For example, North Dakota requires that the majority owner of a pharmacy be a pharmacist, except within hospitals, and in 2018 courts deemed that New Jersey’s state anti-kickback law, the Codey Law, does not permit physician-owned pharmacies. Medicare and Medicaid have also recently interpreted the Stark Law as prohibiting in-house pharmacies owned by independent physician practices from delivering medicines by mail, which many had previously done, and implemented this prohibition in 2023 (Mullangi, 2024).9 These restrictions can be motivated by concerns regarding whether physicians are sufficiently regulated, the influence of financial incentives in prescribing, and pharmacist advocacy. Potential motivators for in-house pharmacy growth in oncology There has been substantial growth in the number of oncology practices operating their own in-house pharmacies in recent years, likely attributable to four key factors (Kanter et al., 2022; Kakani et al., 2024). First, there were an increasing number of high-cost oral chemotherapy agents launched in the last two decades, likely creating a financial incentive for practices to launch pharmacies dispensing these medicines. This was likely caused in part by the introduction of Medicare Part D in 2006, the status of oncology drugs as a protected class, and government reinsurance in the catastrophic coverage phase, muting plans’ cost containment incentives (Blume-Kohout and Sood, 2013; Bennette et al., 2019; Trish et al., 2019). Thus, by 2016 nearly half of chemotherapy spending for Medicare beneficiaries was paid for by Medicare Part D (Kyle et al., 2022). Second, the 340B Drug Discount Program has grown substantially in recent years, increasing the financial incentives for certain hospital-based practices to launch in-house pharmacies (Bach and Sachs, 2018; Kakani et al., 2024). Third, there has been increasing consolidation among oncology practices, including acquisition of practices by health systems and consolidation across independent practices (Nikpay et al., 2018; Milligan et al., 2024). This has likely given more practices sufficient scale to support the fixed costs of launching a pharmacy. Finally, wholesalers such as Cardinal, Cencora, and McKesson have expanded their consulting services and support for practices who want to dispense medications in-house, potentially making it easier for practices to launch pharmacies. Industry reports suggest this may have been partly because wholesalers were able to charge higher drug acquisition prices to practices than to larger, PBM-owned specialty pharmacies (Fein, 2021). Conceptual framework As in other contexts of vertical integration between health care providers, vertical integration between physicians and pharmacies may impact both prices and patient care, which motivates the empirical analysis. These section reviews the theoretical reasons for why vertical integration across health care providers may impact these outcomes and unique features of physician–pharmacy integration that may impact the effect of vertical integration in this context. Prices While the price-increasing effects of horizontal integration among substitutes are well-known, empirical work suggests integration across health care providers that are not substitutes can also increase prices. Physician-hospital integration often increases prices at both the physician group and hospital levels.10 A related literature also shows that integration between hospitals across differing geographic markets can increase prices, despite not being substitutes (Lewis and Pflum, 2017; Dafny et al., 2016). There are several mechanisms that have been theoretically posited for how vertical integration across health care providers can yield price increases. One theory posits that vertically integrated providers achieve higher prices because integrated entities jointly negotiate with payers, such that plans are required to reach an agreement with all or none of the providers within a system. This can result in higher prices if integration increases providers’ bargaining power by increasing the likelihood that patients will switch plans if negotiations fail (Gal-Or, 1999; Peters, 2014). Another possibility is that because vertically integrated entities often form partially exclusive relationships, vertically integrated firms can increase barriers to entry for would-be competitors and thus gain horizontal market power (Gaynor and Vogt, 2000). Finally, larger entities may have data infrastructure and market intelligence advantages that improve bargaining skill (Lewis and Pflum, 2015). The context of physician–pharmacy integration shares certain features with other forms of vertical integration (e.g., physician-hospital integration) that may portend price increases. The shared features include the ability, in theory, to jointly negotiate prices for physician and pharmacy services for MA-PD plans that cover both services. Indeed, hospital-based physicians are known to receive higher prices for physician-administered drugs among commercial plans (Robinson et al., 2021; Xiao et al., 2022), suggesting that a practice’s bargaining power can impact prices of medicines dispensed. Practices can also steer patients to their in-house pharmacy and thus can theoretically prevent the entry of new specialty pharmacies. Finally, the expanded firm could, in theory, gain bargaining skills through market intelligence. However, there are three important points of distinction between physician–pharmacy integration and other forms of integration that may make price increases less likely. First, physician–pharmacy integration occurs via entry of a new pharmacy. This is in contrast with other settings, where integration occurs via acquisition of an existing firm. Thus, integration in this context may be more pro-competitive. Second, the in-house pharmacy typically has limited bargaining power. This is because PBMs, which negotiate prices with pharmacies, are highly consolidated. Furthermore, in-house pharmacies’ main competitors are large specialty pharmacies that are vertically integrated with PBMs. PBMs thus have additional incentive to offer lower reimbursement to in-house pharmacies in order to dissuade them from participating in plan networks and thus to steer patients to their own pharmacies. This is in contrast to other settings, where the acquired entity may have more meaningful bargaining power. Finally, the potential for joint bargaining between medical and pharmacy services is likely lower than in other contexts as medical and pharmacy benefits are often negotiated separately; stand-alone prescription drug plans do not cover physician services, limiting the scope for joint bargaining. Vertical integration between physicians and pharmacies may also impact patient care through the alignment of financial incentives between the physicians and pharmacies and reduced costs of coordination between the physician and pharmacy. These mechanisms are similar to those relevant in other contexts of vertical integration between health care providers including between physicians and hospitals and post acute care facilities. The alignment of financial incentives between the physician and pharmacy may change physician prescribing behavior. This may result in more patients being prescribed high-cost oral cancer treatments and patients being prescribed more expensive regimens. This incentive may be particularly strong in cancer care, where regimens can differ widely in cost (Harrington and Smith, 2008; Chandra and Skinner, 2012; Giordano et al., 2016). Indeed, prior evidence on infused drugs suggests profitability can impact treatment choices (Jacobson et al., 2010; Conti et al., 2012; Bond et al., 2023) and studies of physician drug dispensing in Switzerland have often found dispensing to be associated with higher drug spending (Kaiser and Schmid, 2016; Trottmann et al., 2016; Burkhard et al., 2019). This has the potential to be inefficient if the benefits of additional or costlier chemotherapy are small for the marginal patients. This is analogous to concerns that in other contexts, such as the concern that integration between physicians and hospitals may increase the utilization of hospital-based services (See Post et al. (2018) for review) and that integration between hospitals and post-acute care providers may increase post-acute care use (Konetzka et al., 2018). Conditional on the treatment prescribed, physician–pharmacy integration may also increase and accelerate treatment initiation and improve medication adherence, due both to the alignment of financial incentives between the two entities and reduced costs of coordination. Failure to initiate and delays in initiating treatment are significant concerns for patients prescribed high-cost oral cancer treatment prescriptions; the majority of cancer patients report prior authorization related delays in cancer care (Chino et al., 2023) and up to 30% of patients failing to fill new prescriptions for these medicines (Dusetzina et al., 2022). Integration between the oncology practice and pharmacy could reduce several coordination costs that could both increase or accelerate treatment initiation. For example, integration may reduce practices’ cost of checking whether a patient has filled a prescription by virtue of being observable in the medical record. As many practices have noted, this may also improve adherence after treatment initiation (e.g., The ASCO Post (2017)). Second, integration could reduce the costs associated with determining whether prior authorization is necessary. With an in-house pharmacy, the practice can immediately attempt to fill the prescription and thus determine whether prior authorization is needed quickly, without needing to call the plan or asking patients to return if they are unable to fill the prescription. Third, physician practice based pharmacies may be more convenient for patients by virtue of being co-located with the practice and being familiar. Finally, integration may also increase practice investments in helping patients find financial assistance (Dafny et al., 2022) or monitoring and improving adherence after initiation, due to the alignment of financial incentives between the physician and pharmacy. Data and descriptive statistics Data sources The primary data are a 20% sample of Medicare Fee-for-Service (FFS) and Part D claims from 2006–2019. The FFS claims cover physician, facility, and hospital services provided to FFS patients and include provider identifiers for billing physicians and advanced practice providers (e.g., certain nurse practitioners and physician assistants), tax identification numbers (TIN) for practices, patient diagnoses, services rendered, and payments. Part D claims include drug, prescriber, and pharmacy identifiers, and point-of-sale prices paid to pharmacies for all patients with Medicare Part D, including MA-PD enrollees. See Appendix Section C.1 for details on data cleaning. Medicare Part B claims are linked to the Health Systems and Provider Database (HSPD), which identifies claims from common parent organizations and has been used in prior work (Beaulieu et al., 2023; Ericson et al., 2023). The HSPD combines data from the Medicare Provider Enrollment, Chain, and Ownership System, Irving Levin Mergers & Acquisitions data, public reports, and tax filings to link practice TINs and hospitals owned by common parent organizations over time. All physicians and advanced practice providers billing Medicare FFS are attributed to a parent organization, referred to herein as a “physician organization”, each year based on the plurality of their FFS evaluation and management claims.11 Each year, physician organizations are attributed characteristics. Physician organizations are classified as (i) independent, if it does not own a hospital, or (ii) hospital-linked. Hospital-linked organizations are stratified into those with and without hospitals enrolled in the 340B Drug Pricing Program using the Health Resource & Service Administration Office of Pharmacy Affairs Information System data. Identifying in-house pharmacies Physician organization-operated pharmacies, including dispensaries and licensed pharmacies, referred to collectively herein as “in-house pharmacies”, are identified each year using novel data created, used, and validated in Kakani et al. (2024). These data derive from the National Provider and Plan Enumeration System data, which includes addresses, names, and contact information for individual providers (e.g., physicians) and organizations (e.g., pharmacies, hospitals, and physician practices). The set of pharmacies is first limited to those billing at least $2000 in annual Part D claims, which account for over 99% of spending, and satisfying at least one of three criteria: (a) a shared address with a physician, physician practice, or hospital (b) > 50% of the spending was prescribed by prescribers from one organization, or (c) certification by the Utilization Review Accreditation Committee, an important accreditation for accessing payer contracts, as a specialty pharmacy. Finally, all candidate pharmacies were reviewed manually to identify pharmacies that appeared truly operated by physician organizations through review of pharmacy names, contact information, and websites. As described in Kakani et al. (2024), this list was validated through multiple checks, including cold calling random samples, to ensure virtual completeness. Trends in the use of in-house pharmacies The share of Medicare Part D spending filled at in-house pharmacies grew substantially for high-cost drugs from 2006 to 2019 (Appendix Section D, Appendix Figure 3). The share of oral cancer treatment spending filled by in-house pharmacies grew from 10% to 37%. In other high-cost drug classes such as anti-virals and immunosuppressants, the share of Medicare Part D spending filled by in-house pharmacies grew from 12% to 21% and 5% to 10% respectively. Among other drugs, there was growth in in-house pharmacy use for high-cost drugs (>$10,000 per patient per year) but not other drugs. Finally, within oncology, growth in in-house pharmacy use has been limited to high-cost oral cancer treatments (>$10,000 per patient per year). These findings are consistent with prior work in the Medicare FFS (Kakani et al., 2024) and commercially-insured (Kanter et al., 2022) populations. See Appendix Section C.3, including Appendix Tables A2 and A3, for the oral cancer treatments included. Medical oncology practices with in-house pharmacies Medical oncology practices with in-house pharmacies are then identified each year. Each in-house pharmacy is attributed to the physician organization whose attributed physicians and advanced practice providers prescribe the highest share of spending filled at the pharmacy. Physician organizations are classified as medical oncology practices if they have at least one medical oncologist, with the medical oncology practice being defined as all medical oncologists attributed to the organization; see Appendix Section C.4 for details on identifying medical oncologists. Medical oncology practices are defined as being pharmacy-integrated if at least 15% of spending prescribed by its oncologists on oral cancer treatments is filled via in-house pharmacies attributed to the organization. This threshold is applied because certain organizations, particularly large multi-specialty organizations, often have in-pharmacies that only fill the occasional, rare oncology prescription (e.g., onsite retail pharmacies) but are not equipped to fulfill oncology prescriptions regularly. This threshold thus captures pharmacies that prescribe a meaningful level of oncology prescriptions. This approach to defining medical oncology practices and integration with in-house pharmacies is similar to prior work (Kakani et al., 2024). Annual patient sample for medical oncology practices Each year, the cancer patients treated by each medical oncology practice are then identified. Patients are required to be enrolled in Medicare FFS, Part A, Part B, and Part D continuously or until death, have at least one office visit with a medical oncologist, and have one office visit with a cancer diagnosis during the year. Patients are attributed to medical oncologists based on the plurality of office visits and are attributed to the medical oncology practice of their medical oncologist. Patients are further assigned to one of 22 cancer diagnoses based on which cancer diagnosis appears on the plurality of their office visit claims during the year as in prior work (Keating et al., 2021). See Appendix Section C.5, including Appendix Table A4, for further details on the construction of the annual patient panel for each practice. Characteristics of medical oncology practices Table 1 describes medical oncology practices with and without in-house pharmacies. Practices with in-house pharmacies have substantially more medical oncologists and are less likely to be independent. This is consistent with a model of pharmacy adoption with fixed costs and/or where gross margins increase with size. Practices with in-house pharmacies have fewer dual eligible patients in 2019 and but very similar numbers of patients with hematological cancers, for whom oral chemotherapy is more common. Finally, there is large variation in the share of spending filled in-house among practices with in-house pharmacies. These results are consistent with prior work (Kanter et al., 2022; Kakani et al., 2024). These cross-sectional differences between practices motivate using a matching approach to identify comparable control practices to identify effects of physician–pharmacy integration, as described in Section 5.2. Table 1Characteristics of practices with and without in-house pharmacies.20062019With in-house pharmaciesWithout in-house pharmaciesWith in-house pharmaciesWithout in-house pharmaciesNo. oncologists18.23.026.43.6SD35.56.044.39.0Independent (%)50.585.056.368.3Patients with hematological cancers (%)25.324.329.229.1SD12.221.08.920.7Full dual patients (%)42.637.314.319.6SD23.829.014.324.3Oral cancer treatment spending in-house (%)42.3–53.3–SD23.0–23.3–No. practices9531883731666Note: Table includes practices with at least one medical oncologist. Practices with in-house pharmacies are those where more than 15% of all Medicare Part D spending on oral cancer treatments prescribed by medical oncologists at the practice are filled via the practices’ in-house pharmacies. Patient characteristics derived from patients continuously enrolled in Medicare FFS, Parts A, B, and D who have at least one office visit with a medical oncologist and at least one office visit with a cancer diagnosis during the year. Patients are attributed to medical oncologists based on the plurality of office visits. Medical oncologists are attributed to practices based on the plurality of its evaluation and management visits. Hematological cancer diagnoses are gathered from all Medicare FFS Part A and B claims. Characteristics of practices with and without in-house pharmacies. Note: Table includes practices with at least one medical oncologist. Practices with in-house pharmacies are those where more than 15% of all Medicare Part D spending on oral cancer treatments prescribed by medical oncologists at the practice are filled via the practices’ in-house pharmacies. Patient characteristics derived from patients continuously enrolled in Medicare FFS, Parts A, B, and D who have at least one office visit with a medical oncologist and at least one office visit with a cancer diagnosis during the year. Patients are attributed to medical oncologists based on the plurality of office visits. Medical oncologists are attributed to practices based on the plurality of its evaluation and management visits. Hematological cancer diagnoses are gathered from all Medicare FFS Part A and B claims. Stacked event study design A stacked event study design is then used to evaluate the effects of physician–pharmacy integration on outcomes related to point-of-sale prices and patient care. Identifying physician–pharmacy integration events The analysis focuses on 204 medical oncology practices integrating with in-house pharmacies between 2009 to 2016. These integration events are intended to capture events where a medical oncology practice discontinuously transitions from having very limited or no integration to having substantial integration with an in-house pharmacy, measured by the share of the medical oncologists’ prescriptions for oral cancer treatments filled in-house, and excluding events co-occurring with significant expansions or practice acquisitions. Thus the sample includes events where physician organizations are launching in-house pharmacies for the first time and events where physician organizations are significantly expanding existing pharmacy services to regularly supply oral cancer treatments and to be clinically integrated with the medical oncology practice. This is similar to the notion of integration in prior work studying the effects of fragmented care delivery (Agha et al., 2019; Ericson et al., 2023; Agha et al., 2023). This approach confers two main advantages, relative to focusing on events where a medical oncology practice simply goes from not having to having an in-house pharmacy in any capacity. First, this approach focuses on events that meaningfully change clinical integration between the medical oncology practice and pharmacy.12 Second, this approach permits studying the effects of physician–pharmacy integration at physician organizations, particularly large, multi-specialty health systems, that often have pre-existing non-oncology pharmacies filling the occasional, rare oncology prescriptions (e.g., onsite retail pharmacies), but which choose to launch a full specialty pharmacy service integrated with their medical oncology practice. Given substantial growth in the share of oncologists employed by hospitals and health systems over this period, inclusion of these events is increasingly important (Nikpay et al., 2018). Integration events are therefore defined through a multi-step process. Preliminary events are identified by searching for practices that have a four year period (t−3,t−2,t−1,t=0) in which the practice employs at least one medical oncologist prescribing in Medicare Part D in all four periods and t=0 is the first year in which the practice is pharmacy-integrated. In the main specification, practices are categorized as pharmacy-integrated if > 15% of the oral cancer treatment spending prescribed by the practices’ medical oncologists in Medicare Part D is filled at in-house pharmacies. However, results are shown to be insensitive to weaker thresholds (e.g., 5% or 10%). Given the imprecise timing of integration events, I also demonstrate the insensitivity of results to defining t−1 as the event year, dropping observations in year t−1, and dropping observations in year t=0. Finally, results are shown to be insensitive to using the subset of treatment practices that had no in-house pharmacy filling any oral cancer treatments prior to t−1. The list of preliminary events is refined to derive the final list of practices integrating with in-house pharmacies. First, all pharmacies linked to preliminary events were manually reviewed and a small number of erroneous linkages were corrected; these cases likely reflect physicians affiliated with multiple practices or external patients filling prescriptions at the practice. Second, events where the number of medical oncologists at the practice increases by 20 or more, or where the organization changes 340B affiliation, between t−1 and t+1 were excluded, as these practices are likely undergoing significant co-occurring changes that could confound results. Finally, the sample is limited to practices that exist and employ at least one medical oncologist prescribing in Medicare Part D for each of the 3 years following the event (t+1,t+2,t+3). This excludes practices that are acquired or closed in the 3 years following the event and events occurring between 2017–2019. Appendix Section E, Appendix Figure 4 illustrates this procedure and the number of oncology practices integrating with pharmacies over time.13 Identifying control practices For each integration event between 2009–2016 (treatment group), up to ten control practices are identified in the main specification. For each treated practice, the set of control practices are drawn from the set of candidate control practices that meet all the same inclusion criteria as treated practices, except they are not meaningfully integrated with a pharmacy for any years from t−3 to t+3. Furthermore, only practices that either never underwent a pharmacy integration event (“never-treated practices”) or practices that underwent a pharmacy integration event five or more years after the treatment practice (“late-treated practices”) are considered candidate controls. Other practices undergoing pharmacy integration events at different times are not included as candidate controls because including these observations as controls can contaminate estimated treatment effects (Goodman-Bacon, 2021; Callaway and Sant’Anna, 2021; Sun and Abraham, 2021). In the main analysis, treatment practices are matched to control practices based on the number of medical oncologists attributed in the year prior to the event using coarsened-exact matching. This yields 8.2 control practices per event on average, with practices permitted to serve as controls for multiple treatment practices. Nonetheless, results are robust to alternative matching strategies including using all eligible control practices (no matching) or using only late-treated practices as controls. Appendix Section E, Appendix Figure 5 illustrates differences in cross-sectional characteristics between treatment and matched control practices in the year prior to the event across three matching approaches: matching on size (main analysis), no matching, and using late-treated practices only. While none of these strategies is able to eliminate all differences between treatment and control practices, this is not required for identification as discussed in Section 5.4. Model specification The primary effect estimates are derived from linear models of the form in Eqs. (1) and (2) using data from treatment and control practices in the five years before and after the integration event in the treatment practices.14 Here, Yipct refers to an outcome Y for patient index event i attributed to practice p in experiment c in year t. Experiments are defined as the set of treatment practices undergoing pharmacy integration events in a given year within coarsened buckets of practice size (1, 2–3, 4–7, 8–15, or 16+ oncologists) and the set of control practices attributed to them. There are, thus, 40 possible experiments (8 event years ×5 coarsened buckets of practice size). Then, γct is a set of fixed effects for each experiment-calendar year and controls for secular trends within each experiment, ξpc is a set of fixed effects for each practice-experiment, Xipct is a set of covariates that vary by outcome (discussed in Section 6), eventc refers to the event year associated with experiment c, yeart refers to the calendar year t, treatpc is an indicator for whether practice p in experiment c is in the treatment group, and postct is an indicator for whether year t for the experiment c is after the integration event. Standard errors are clustered by practice. In Eq. (1), the key coefficients are τj, which reflect the relative growth in Yipct associated with the treatment relative to the control group each year before and after the integration event, relative to baseline group differences in the year prior to the event. In Eq. (2) the key coefficient is δ and reflects the average effect of integration. Including dummy variables for each experiment interacted with time and practice assures that estimates represent an average effect within homogenously timed experiments and thus do not suffer from identification issues associated with two-way fixed effects models with heterogeneous treatment timing (Goodman-Bacon, 2021; Callaway and Sant’Anna, 2021; Sun and Abraham, 2021). This is also similar to approaches taken in prior work (Brot-Goldberg et al., 2023). The coefficient δ in Eq. (1) represents an aggregate effect of physician–pharmacy integration across all patients. However, it should not be assumed that this effect derives entirely from the effect on patients who use the in-house pharmacy. This is because it is possible that physician–pharmacy integration impacts outcomes for patients who use external pharmacies. For example, practices may develop a workflow wherein the in-house pharmacy staff coordinate care for all patients prescribed oral cancer drugs, including determining if prior authorization is needed and monitoring medication adherence issues, even if the patient ultimately uses another pharmacy. Identification assumptions The key assumption needed to impart a causal interpretation to this analysis is that treatment group outcomes would have trended in parallel to the control group in the absence of the integration event, and several observations are made to develop confidence in this assumption. First, outcomes in treatment and control groups exhibit parallel trends for each outcome prior to the event, suggesting that these practices would have continued on a similar trajectory in the absence of the event. Second, in Appendix Section F Appendix Figures 6–8, a series of placebo tests illustrate that observable patient and practice characteristics are not changing at the same time as the integration event. Finally, while it is possible that patients are changing along unobserved dimensions, the likelihood that this explains the results is gauged by evaluating the sensitivity of results to the exclusion of observable covariates (Oster, 2019). Ensuring events reflect meaningful integration The identified integration events are confirmed to mark a meaningful and sustained change in the practices’ level of pharmacy integration. This is verified by showing that the share of spending in Medicare Part D on oral cancer treatments prescribed by oncologists in the practice increases substantially and sustainably after the integration event year (Fig. 1, Panel A). This increase is considerably stronger for spending on high-cost than low-cost oral cancer treatments, validating the focus on outcomes related to high-cost oral cancer treatments. I also use the difference-in-difference framework outlined in (1) and (2) to confirm that these increases are statistically significant (Fig. 1, Panel B). Specifically, I estimate versions (1) and (2) where the outcome is instead of the form Ypct where Y is the outcome for a practice p in experiment cohort c in year t, and there are no patient controls Xipct. The outcome is the share of Medicare Part D spending on high-cost oral cancer treatments prescribed by oncologists in the practice that occurs at in-house pharmacies. I also repeat this analysis for low-cost oral cancer treatments. I confirm that integration is associated with an 37.7 (95% CI: 34.8, 40.7) percentage point (pp) increase in the share of spending captured in-house for high-cost oral cancer treatments but only a 9.0 pp (95% CI: 7.1, 10.8) increase in the analogous quantity for low-cost oral cancer treatments. Notably, the share of spending filled in-house does not reach 100%. This is consistent with variable inclusion of in-house pharmacies in Part D networks, variable stocking of medicines at the in-house pharmacy, and patients retaining the option to fill externally. Some patients may prefer to fill externally if, for example, their in-house pharmacy does not deliver medications by mail. Mean share of Medicare Part D spending on oral cancer treatments filled in-house in treatment and control practices before and after integration event. Note: Panel A plots the mean share of Medicare Part D spending on oral cancer treatments prescribed by medical oncologists filled at practices’ in-house pharmacies each year before and after the integration event. The share of Medicare Part D spending on oral cancer treatments filled in-house is first estimated for each practice, each year and then averaged across practice-years. Panel B reflects regression adjusted estimates of τj from the version of Eq. (1) described in Section 5.5 and 95% confidence intervals. The overall difference-in-difference (DiD) effect reflects an estimate of δ from the version of Eq. (2) described in Section 5.5. High-cost oral cancer treatments are those drug molecules with greater than $10,000/year median annual cost over the study period. Low-cost oral cancer treatments are all other oral cancer treatments. See Appendix Section C.3, including Appendix Tables A2 and A3, for the oral cancer treatments included. Outcomes The effects of physician–pharmacy integration are evaluated on outcomes related to point-of-sale prices and patient care. Patient care outcomes reflect three key dimensions: treatment initiation (including likelihood and timeliness), adherence following initiation, and substitution to alternative treatments. Given heterogeneity in outcomes and to maximize power, the unit of analysis and covariates Xipct, per Eqs. (1) and (2), vary by outcome as described herein. The attribution of patients’ to oncologists also differs across outcomes, as explained herein. Across outcomes, patients are attributed to the practice with which their oncologist has the most office visits in the calendar year. Further details on sample construction can be found in Appendix Section C.5 including Appendix Table A4. Point-of-sale prices The unit of analysis is each Part D claim for high-cost oral cancer treatments prescribed by oncologists at in-sample practices. Claims are thus attributed to the prescribing oncologist. The covariates Xipct are drug-year-plan fixed effects, with drugs defined by National Drug Codes. Likelihood of treatment initiation The unit of analysis is patient-years, with the sample defined as in Table 1. As in Table 1, patients are attributed to the oncologist associated with the plurality of their office visits. The outcome is whether the patient had a “first-time fill” of a high-cost oral cancer treatment, defined as a new initiation of at least one molecule for the first time in Medicare Part D.15 All fills, regardless of pharmacy used are considered. Thus, evaluating the impact of integration on this measure evaluates the degree to which integration results in market expansion on the extensive margin. The covariates Xipct are 5-year age bins, sex, race and ethnicity, Charlson comorbidity index with Klabunde modification (Klabunde et al., 2007), cancer diagnosis × 5-year period, metastatic disease × 5-year period, full dual, partial dual, disabled, urban residence, and full and partial low-income subsidy. Importantly, we cannot observe prescriptions that are prescribed but not filled and thus cannot evaluate separate the effects of integration on new prescriptions versus abandonment of existing prescriptions. Time-to-fill during treatment initiation The unit of analysis is each first-time fill of a high-cost oral cancer treatment prescribed, with observations being attributed to the prescribing medical oncologist. The sample is limited to patients enrolled in Medicare FFS Parts A, B, and D for the six months prior who also had an evaluation and management visit with the prescriber in the six months prior.16 Time-to-fill is then estimated as the number of days between the first-time fill and the patients’ last evaluation and management visit with the prescribing oncologist. Importantly, this is not an exact measure of the time-to-fill because claims data do not reveal when a prescription is written, but it is a reasonable proxy, given that new prescriptions typically require a consult beforehand, during which a prescription is written.17 The covariates Xipct include 5-year age bins, sex, race and ethnicity, full dual, partial dual, disabled, urban residence, full low-income subsidy, partial low-income subsidy, and molecule fixed effects. Adherence The unit of analysis is each first-time fill for a high-cost oral cancer treatment prescribed, with observations being attributed to the prescribing medical oncologist. Patients are required to be continuously enrolled in Medicare Part D for the 6 months prior and 12 months following or until death and have at least two fills of the drug during the year following the first fill. Medication adherence is measured as the medication possession ratio, or the proportion of days between the first and final fill within the first year, that the patient had the drug.18 The covariates Xipct include 5-year age bins, sex, race and ethnicity, full dual, partial dual, disabled, urban residence, full low-income subsidy, partial low-income subsidy, and molecule fixed effects. The medication possession ratio therefore measures adherence during the period between the first and the last fill during the year following treatment initiation, but does have limitations. First, it does not measure non-adherence related to early discontinuation as one cannot easily differentiate discontinuation due to non-adherence versus an intentional discontinuation. Second, the measure only considers whether the patient is in possession of a medication and does not assess whether the medication is actually taken. Finally, the measure assumes that patients should be taking the drug continuously between the first and final fill. This is generally a reasonable assumption for high-cost oral cancer drugs, because patients are not typically prescribed the same drug for multiple lines of therapy. However, there may be some exceptions for patients experiencing significant side effects or other complications, who may be asked to pause medication intentionally. Substitution Finally, the effect of physician–pharmacy integration on substitution to alternative treatment regimens among high-cost oral cancer treatments is evaluated. Substitution can include substitution to higher cost regimens within high-cost oral cancer treatments (e.g., longer duration or higher cost molecules) and substitution away or towards alternative or complimentary therapies (e.g., low-cost oral cancer treatments and infused Part B chemotherapies). In the main analysis, the effects of integration are evaluated on an aggregate measure of regimen costliness: total spending on high-cost oral cancer treatments. The core patient cohort and covariates Xipct are as described in Section 6.2, except the analysis is limited to patients with at least one prescription for high-cost oral cancer treatments and thus reflects the intensive margin of utilization, complementing the outcome described in Section 6.2. In supplementary analysis, other forms of substitution are evaluated including effects on both the extensive and intensive margins of low-cost oral cancer treatments and infused Part B chemotherapy utilization, including for the subset of cancers with specific opportunities for substitution between high-cost oral Part D and infused Part B chemotherapies. Supplemental analyses also consider a specific case study in which substitution between high-cost oral cancer treatments may be particularly plausible, namely the use of 1st generation (comparatively lower cost) vs. 2nd or 3rd generation (comparatively higher cost) tyrosine kinase inhibitors as first-line therapy for chronic myeloid leukemia (Nguyen et al., 2020). Price effects Following the integration event, point-of-sale prices for high-cost oral cancer treatments prescribed to patients of the treatment practice decrease by 0.28% (95% CI: 0.08%, 0.47%), as illustrated in Fig. 2 (Panels A & B). These results are generally insensitive to a range of specifications using alternative control groups and event definitions and using a balanced panel (Appendix Section G, Appendix Figure 9). Impact of physician–pharmacy integration on point-of-sale prices paid for high-cost oral cancer treatments in Medicare Part D. Note: Panel A reflects the unadjusted trends in log(prices) before and after the integration event, after subtracting the mean value within drug-plan-year combinations. Panel B reflects regression adjusted estimates of τj from Eq. (1) and 95% confidence intervals. The overall difference-in-difference (DiD) effect reflects an estimate of δ from Eq. (2). Panel C stratifies the treatment claims after the event from Panel A between claims filled at the practices’ in-house pharmacy relative to external pharmacies. Panel D reflects the estimates of v (Treatment × Post) and η (Treatment × Post × In-house) respectively from Eq. (3). Regression estimates in Panels B and D adjust for practice-experiment fixed effects, experiment-year fixed effects, and drug-year-plan fixed effects. For this analysis, drugs refer to National Drug Codes (NDCs). Standard errors are clustered by practice. The overall effect captures effects related to lower prices at in-house pharmacies relative to external prices and any decreases in prices at external pharmacies in response to the integration event. Eq. (3) is used to decompose these mechanisms. Eq. (3) is a variant of Eq. (2) where inhouseipct is an indicator for whether the prescription was filled in-house and equals 0 for control practices and treatment practices prior to the event. Thus, v and η reflect the change in price after the event attributable to fills occurring externally and internally respectively. Notably, Eq. (3) is only used for the analysis of prices but not patient care outcomes for two reasons. First, Eq. (3) cannot be used for analyses of certain patient outcomes (e.g., treatment initiation) where many patients do not use any pharmacy. For outcomes (e.g., time-to-fill, adherence) where patients do use a specific pharmacy, the decision on whether to use the in-house or external pharmacies may be correlated with unobservable patient characteristics that could impact patient outcomes. For example, if an in-house pharmacy does not deliver medication by mail, then patients who are aware that mail-order pharmacies are an option and are more comfortable voicing their preference to use the mail-order pharmacy may be less likely to fill in-house. These patients may also be the patients who are more likely to be adherent. While the degree to which there is selection into in-house pharmacy use on unobservable factors is unknowable, supplemental analysis reveals that patients at treated practices filling new prescriptions at in-house pharmacies do differ in certain observable characteristics versus those filling at external pharmacies (Appendix Section G, Appendix Table A5). Specifically, patients at treatment practices who are Black, are partial dual-eligibles, or have more comorbidities are significantly more likely to using an in-house pharmacy. This suggestive that selection into in-house pharmacy use on unobservable factors may be possible. The aforementioned two issues are not a concern for prices. This is because an analysis of prices requires pharmacy utilization and patient selection is not a concern. Patient selection is not a concern because prices for a given drug are determined not by patient characteristics but only the contract between pharmacies and plans. Therefore Eq. (3) applied to prices is more useful in disentangling whether effects are driven by changes for patients using the in-house pharmacy or for patients using external pharmacies than a similar analysis would be for patient outcomes. As illustrated in Fig. 2 (Panels C & D), the effect on prices is exclusively driven by 1.00% (95% CI: 0.72%, 1.29%) lower prices at in-house pharmacies, whereas there is no detectable effect on prices at external pharmacies. These event study differences are approximately equivalent to the cross-sectional differences between prices at in-house pharmacies versus external pharmacies in Kakani et al. (2024). As illustrated in supplementary materials, this reduction in price associated with in-house pharmacies persists even for MA-PD plans, where joint bargaining over physician and pharmacy services may be relevant, and at large physician organizations with > 25% market share, which may have the greatest bargaining power. The negative association between prices and in-house pharmacies is weaker and not statistically different from 0 when evaluating point-of-sale prices at in-house pharmacies owned by large organizations in MA-PD plans, but meaningful price increases can be ruled out even in this case (Appendix Section G, Appendix Figure 10). These results suggest that the lack of price increases at in-house pharmacies is not driven by limited joint bargaining over medical and pharmacy benefits. Rather, lower prices may appear because in-house pharmacies individually have limited bargaining power compared to PBMs, which are highly consolidated. Moreover, the specialty pharmacy market is dominated by large, PBM-owned pharmacies largely operating through mail. Thus, PBMs may have additional incentive to suppress pricing at in-house pharmacies as much as possible, subject to the constraints of “Any Willing Pharmacy” regulations, in order to deter in-house pharmacies from joining their pharmacy networks. This is because if in-house pharmacies opt out of the plan’s network, the PBM may be more easily able to steer patients to their own pharmacies. Government reports provide some support for this hypothesis, by showing that large PBMs pay external pharmacies less than their own pharmacies for certain specialty medications, and PBMs often offer unaffiliated pharmacies challenging contract terms (FTC, 2024). Patient care effects Physician–pharmacy integration also increases and accelerates treatment initiation among high-cost oral cancer treatments (Fig. 3). Specifically, integration increases the likelihood that a patient has a first-time fill of a high-cost cost oral cancer treatment by 0.19 percentage points (p.p.) (95% C.I: 0.06, 0.33). This represents an increase in the likelihood of a first-time fill by 6.2% (95% C.I.: 1.9, 10.4) relative to the sample mean of 3.13 p.p. in the control group (Fig. 3 Panel B). Similarly, integration increases the likelihood that a patient has a first-time fill of a high-cost oral cancer treatment within 5 days of the last physician consult by 9.17 p.p. (95% CI: 6.11, 12.22). This represents an increase of 21.8% (95% C.I.: 14.5, 29.1) relative to the sample mean 42.00 p.p. in the control group (Fig. 3 Panel D). Impact of physician–pharmacy integration on likelihood and timeliness of treatment initiation for high-cost oral cancer treatments in Medicare Part D. Note: Panel A reflects the unadjusted share of patients with first-time fills for a high-cost oral cancer treatment in Medicare D; a first-time fill refers to an instance in which a specific high-cost oral cancer treatment drug, defined by molecule name, is filled by a patient for the first time in Medicare Part D. Panel C reflects the unadjusted share of first-time fills for high-cost oral oncology treatments in Medicare Part D (patient-molecule combinations) each year that occur within 5 days of the last physician consult with the prescribing oncologist. Panels B and D reflect regression adjusted estimates of τj from Eq. (1) and 95% confidence intervals. The overall difference-in-difference (DiD) effect reflects an estimate of δ from Eq. (2). Panel B and D both adjust for practice-experiment fixed effects, experiment-year fixed effects, five-year age bins, sex, race and ethnicity, full dual, partial dual, disabled, urban/rural, full low-income subsidy enrolled, and partial low-income subsidy enrolled. Panel B additionally controls for Charlson comorbidity index, cancer diagnosis and metastatic disease; cancer diagnosis and metastatic disease are interacted with the multi-year time windows for 2006–2010, 2011–2015, 2016–2019. Panel D additionally controls for molecule fixed effects. Standard errors are clustered by practice. However, physician–pharmacy integration does not improve medication adherence (Fig. 4). Rather, integration appears associated with a 1.42 p.p. (95% CI: −2.23, −0.61) reduction in the medication possession ratio. There are two competing theories for the modest reduction. The first is that integrated pharmacies are less skilled in monitoring and ensuring adherence relative to larger pharmacies suggesting reduced welfare. The second is that the marginal patients serviced by integrated pharmacies are more likely to be non-adherent. Indeed, if the marginal patients have a medication possession ratio 1.0 S.D. below the mean, this would be sufficient to explain the effect. In this case the integrated pharmacies may not be less skilled and welfare may not be harmed for infra-marginal patients. Nonetheless, a 1.42 p.p. reduction in the medication possession ratio is unlikely to be clinically significant even if driven by inframarginal patients. At the same time, it is exceedingly unlikely that lower adherence of the marginal patients is masking important improvements in adherence for inframarginal patients; marginal patients would need to have a medication possession ratio 4.72 S.D. below the mean to mask a 3.0 p.p. improvement in adherence for inframarginal patients. Impact of physician–pharmacy integration on medication adherence upon treatment initiation for patients receiving high-cost oral cancer treatments covered by Medicare Part D. Note: Panel A reflects trends in unadjusted average medication possession ratio for patients in their first year of taking a specific drug. Panel B reflects regression adjusted estimates of τj from Eq. (1) and 95% confidence intervals. The overall difference-in-difference (DiD) estimator reflects an estimate of δ from Eq. (2). Regression estimates adjust for practice-experiment fixed effects, experiment-year fixed effects, drug molecule fixed effects, five-year age bins, sex, race and ethnicity, full dual, partial dual, disabled, urban/rural, full low-income subsidy, and partial low-income subsidy. Standard errors are clustered by practice. Physician–pharmacy integration also does not result in higher spending on high-cost oral cancer treatments among recipients (Fig. 5), suggesting limited substitution to more expensive regimens. Supplementary analyses also do not show evidence of substitution to or away from low-cost oral Part D cancer treatments or infused Part D chemotherapies, including when this analysis is limited to diagnoses with specific opportunities for substitution. Among chronic myeloid leukemia patients initiating treatment with a tyrosine kinase inhibitor, there was also no increase in propensity to use higher-cost, 2nd or 3rd generation tyrosine kinase inhibitor, though this analysis had important power constraints (Appendix Section G, Table A6). Impact of physician–pharmacy integration on high-cost oral cancer treatment spending among recipients in Medicare Part D. Note: Panel A reflects trends in unadjusted average spending on high-cost oral cancer treatments among patients receiving high-cost oral cancer treatments. Panel B reflects regression adjusted estimates of τj from Eq. (1) and 95% confidence intervals. The overall difference-in-difference (DiD) estimator reflects an estimate of δ from Eq. (2). Regression estimates adjust for practice-experiment fixed effects, experiment-year fixed effects, cancer diagnosis, metastatic disease, five-year age bins, sex, race and ethnicity, full dual, charlson score, partial dual, disabled, urban/rural, full low-income subsidy enrolled, and partial low-income subsidy enrolled. Cancer diagnosis and metastatic disease are interacted with the multi-year time windows for 2006–2010, 2011–2015, 2016–2019. Standard errors are clustered by practice. These results are generally insensitive to various sensitivity analyses using alternative matching strategies, alternative event definitions, no patient controls, or using a balanced panel of observations (Appendix Section G, Appendix Figures 11–13). There are also no detectable effects on time-to-fill and adherence for low-cost oral cancer treatments (Appendix Section G, Table A6). This provides further confidence that the observed pattern of effects is not driven by simultaneous changes in other aspects of practice operations, independent of the pharmacy integration Mechanisms These results suggest that physician-pharmacy integration primarily impacts the likelihood and speed of treatment initiation, for which several mechanisms are possible. It is possible that the profit incentive created for the physician organization increases prescribing. It is alternatively possible that integration streamlines processes that increase the likelihood and speed of treatment initiation conditional on prescriptions through several means. First, practices integrated with pharmacies may invest in helping patients find financial assistance programs.19 Practices may be more equipped than external pharmacies to support patients in finding and applying for opportunities, given their understanding of patients’ clinical needs, and practices with in-house pharmacies may have greater incentives to do so. Second, it is possible that practices with pharmacies are able to speed prior authorization. The set of drugs under prior authorization varies by plan and over time, posing a challenge in determining whether prior authorization is necessary. Without an in-house pharmacy the practice may call the health plan or pharmacy or ask the patient to determine whether prior authorization was required. Practices with pharmacies, however, could more quickly test whether prior authorization is necessary by attempting to fill a prescription without prior authorization. Third, it is possible that practices are better able to follow-up with patients who fail to fill because integration allows the practice to observe whether the patient has filled a prescription. Fourth, it is possible that the in-house pharmacy is simply more convenient for patients than an external specialty pharmacy by virtue of being co-located with the practice, familiar, and/or approachable. Finally, it is possible that the ability for integration to improve coordination is mediated by practice size, with larger practices more able to coordinate care. Methods The relative plausibility of these mechanisms is evaluated through a series of tests. First, I stratify results by two key organization characteristics: organization type (system without a 340B hospital, system with a 340B enrolled hospital, and independent) and practice size (above versus below median20 number of oncologists). Heterogeneity by organization type may be indicative of whether financial incentives are driving effects. Specifically, prescription fills are more profitable for organizations enrolled in 340B. At the same time, oncologists in independent practices are more likely to have their personal income tied to chemotherapy profits (Malin et al., 2013). Practice size may also mediate effects if larger practices are more likely to have the scale necessary to achieve coordination benefits. I do this by estimating Eq. (4). Here Yiphct refers to an outcome Y for patient i in attributed to practice p of organization type (system without a 340B hospital, system with a 340B enrolled hospital, and independent) h in experiment c in year t. Thus γhct is an experiment-organization type-time fixed effect,21 ξphc is a practice-experiment fixed effect, Xihpct is a set of covariates, treatpc is an indicator for whether the practice is treated, postct is an indicator for whether the year is after the event, and system340Biph is an indicator for whether the practice is affiliated with a 340B hospital, independentiph is an indicator for whether the practice is independent, and abovemedianip is an indicator for whether the practice has an above median number of oncologists in the year prior to the event. Thus α1 reflects the degree to which integration impacts system-based practices below median practice size without a 340B affiliation. Meanwhile α2, α3, and α4 are the additional impacts observed for system-based practices with a 340B affiliation, independent practices, and practices above median practice size respectively. Second, to evaluate the likely role of improved financial support, results are stratified for patients receiving full subsidies in the low-income subsidy program compared to those that are not. If effects are stronger among patients not receiving low-income subsidies, this may suggest integration causes practices to better support patients in finding financial assistance. This is done by estimating Eq. (5). This equation reflects a variant of Eq. (4) where the outcome is Yigpct refers to an outcome Y for patient i in group g (with or without low-income subsidy) attributed to practice p in experiment c in year t. Thus γgct is an group-experiment-time fixed effect, ξgpc is a group-practice-experiment fixed effect, Xigpct is a set of covariates, treatpc is an indicator for whether the practice is treated, postct is an indicator for whether the year is after the event, and nolowincomeig is an indicator for the patient not being eligible for the low-income subsidy. Thus α1 reflects the degree to which integration impacts patients receiving low-income subsidies, and α2 reflects whether there is an additional impact for patients who are not enrolled in the low-income subsidy. Third, to evaluate the role of prior authorization, results are stratified for drugs under versus not under prior authorization. For evaluating heterogeneity in the impact of integration on first fills, this was done by stacking two versions of the data, one where the outcome was defined as first fills of drugs under prior authorization and another where the outcome was defined as first fills of drugs not under prior authorization. For evaluating heterogeneity in the impact of integration on time-to-fill, prior authorization was assigned based on whether the first fill required prior authorization. For both outcomes, Eq. (6) was estimated. This equation reflects a variant of Eq. (5) where the outcome is Yigpct refers to an outcome Y for patient i in group g (with or without prior authorization) attributed to practice p in experiment c in year t. Thus γgct is an experiment-group-time fixed effect, ξgpc is a group-practice-experiment fixed effect, Xigpct are patient covariates, treatpc is an indicator for whether the practice is treated, postct is an indicator for whether the year is after the event, and priorauthig is an indicator for whether the observation is associated with prior authorization. Thus α1 reflects the degree to which integration impacts outcomes for drugs not under prior authorization, and α2 reflects whether there is an additional impact when drugs are under prior authorization. For each heterogeneity test, several sensitivity analyses are also performed. These include evaluating the sensitivity of results to alternative matching strategies, alternative event definitions, no patient controls, or using a balanced panel of observations is also evaluated (Appendix Section G, Appendix Tables A7–A12). Finally, to evaluate the possibility that practices are better able to follow-up with patients who have long delays in filling prescriptions, I evaluate how the distribution of time-to-fill for high-cost oral cancer treatments changes before and after the integration event. This is done by estimating separate stacked event study regressions, following Eq. (2), for each possible value of time-to-fill ranging from 0 to 28+ days. Results The effects of integration on first-fills and the time-to-fill are not ostensibly heterogeneous by organization type and are not stronger in larger practices (Table 2, Models 1 and 4). This is consistent with a limited role for organizational financial incentives (e.g., via the 340B program) or individual oncologist financial incentives in driving the increased likelihood and speed of first-time fills for high-cost oral cancer treatments. While there are power limitations in the ability to detect differences by organization type, the analysis can rule out the possibility that effects on first fills are limited to practices affiliated with 340B hospitals or independent practices. This is because effects persist and are significant for system-based practices unaffiliated with 340B hospitals. These results are robust across alternative specifications (Appendix Table A7 and A10). The results also rule out stronger effects in larger practices. On the contrary, results suggest effects on first fills may be slightly weaker in larger practices, though this result is marginally significant in the main specification and statistically insignificant in most sensitivity analyses (Appendix Table A7). Table 2Tests for heterogeneous effects of physician–pharmacy integration on increased and faster first-time fills.First fill of high-cost drugFirst fill within 5 days of last consult(1)(2)(3)(4)(5)(6)Post × treat.0056.0003.0004.0579.0839−.024095% CI(.0009, .0104)(−.0024, .0030)(−.0004, .0011)(−.0689, .1848)(.0391, .1288)(−.1105, .0625)Post × treat × 340B−.0027.005895% CI(−.0070, .0017)(−.1207, .1323)Post × treat × independent−0.0002. 045695% CI(−.0045, .0041)(−.0774, .1687)Post × treat × above median practice size−.0032−.005395% CI(−.0062, −.0001)(−.0617, .0723)Post × treat × non-LIS.0025.008095% CI(−.0004, .0053)(−.0434, .0594)Post × treat × prior authorization.0012.103295% CI(−.0007, .0030)(.0111, .1953)Observations (N)3 138 8663 138 8626 277 73288 65988 50787 131Note: Models evaluate how the main effects in Fig. 3 Panels B and D vary by key factors. Models 1 and 4 reflects result from estimating Eq. (4), models 2 and 5 reflects result from estimating Eq. (5), and models 4 and 6 reflects result from estimating Eq. (6). Models 1–3 adjust for the same patient covariates Xihpct as in Fig. 3 Panel B. Models 4–6 adjust for the same patient covariates Xihpct as in Fig. 3 Panel D. Less than 2% of observations are dropped in each regression due to having no variation in the outcome within the fixed effects. Standard errors are clustered by practice. Tests for heterogeneous effects of physician–pharmacy integration on increased and faster first-time fills. Note: Models evaluate how the main effects in Fig. 3 Panels B and D vary by key factors. Models 1 and 4 reflects result from estimating Eq. (4), models 2 and 5 reflects result from estimating Eq. (5), and models 4 and 6 reflects result from estimating Eq. (6). Models 1–3 adjust for the same patient covariates Xihpct as in Fig. 3 Panel B. Models 4–6 adjust for the same patient covariates Xihpct as in Fig. 3 Panel D. Less than 2% of observations are dropped in each regression due to having no variation in the outcome within the fixed effects. Standard errors are clustered by practice. The effect of physician-pharmacy integration on first-time fills is trending towards being stronger in non-subsidized patients suggesting that better help finding financial assistance may be a relevant mechanism for increasing first-time fills (Table 2, Model 2). However, these results should be viewed with caution as power is limited in this analysis, and thus the 95% confidence interval does not rule out a null effect. Similarly, the time-to-fill is not heterogeneous by low-income subsidy status, which might be expected if practices were indeed better in helping patients find financial assistance (Table 2, Model 4). These results are generally consistent across specifications (Appendix Tables A8 and A11) The hypothesis with the strongest support from these analyses is that in-house pharmacies may perform faster prior authorization. First-time fills for high-cost oral cancer treatments are faster only when the drug is under prior authorization (Table 2, Model 6). This result also suggests that convenience is not the principal factor driving faster fills, as this would cause faster fills even in the absence of prior authorization. The effect of integration on the likelihood of first-time fills are is also trending towards being higher for drugs under prior authorization (Table 2, Model 3). These results are consistent across specifications (Appendix Tables A9 and A12). Finally, the effect of physician–pharmacy integration on the distribution of days between first-fills and last office visits is also consistent with faster prior authorization (Fig. 6). There are reductions in the likelihood of patients receiving first-time fills 7–14 days after their last office visit, which may be a typical timeline for prior authorization. Meanwhile, physician–pharmacy integration does not decrease long delays (> 14 days) suggesting that follow-up with non-fillers is not a primary mechanism. Predicted distribution of days between first-time fill for high-cost oral cancer treatment and last consult in treated and untreated practices. Note: This figure reflects the predicted distribution of days between a first-time for high-cost oral cancer treatment and last consult in treated practices integrating with pharmacies versus untreated, control practices. The figure reflects the results of 29 regressions of the form outlined in Eq. (2) in which the outcomes vary and are indicators for whether the number of days between the first-time fill and last consult is equal to values from 0 to 27 or 28+. Likelihoods are predicted assuming the mean value of all covariates. Regression estimates adjust for the same set of patient covariates Xihpct as in Fig. 4 Panel D. Welfare analysis The welfare effect of physician–pharmacy integration has several components. The most important effect is the increase in the likelihood of first-time fills for high-cost oral cancer treatments. This is likely to be net positive from a patient health perspective,22 but the benefits relative to costs are highly uncertain. Secondary effects include any clinical value of faster first-time fills, lower adherence (if driven by infra-marginal patients) and a potential reduction in patient or provider hassle costs through smoother prior authorization.23 The integration also generates two transfers. The primary transfer is from external pharmacies to physician practices for profits on infra-marginal utilization. Secondarily, a reduction in prices is a transfer from external pharmacies to Medicare Part D Plans, tax-payers, and patients, who share the cost of drugs. Any government savings from this transfer could, in theory, result in a positive fiscal externality due to the cost of raising public funds. One approach to considering welfare is to consider the cost-effectiveness per quality-adjusted life year (QALY) of marginal uses of high-cost oral cancer treatments. To probe whether marginal use of high-cost oral cancer treatment were likely to be low-value, three elements of treatment effect heterogeneity were examined: (1) end-of-life vs. overall utilization, (2) use by older & sicker vs. younger & healthier patients, and (3) use by patients with metastatic vs. non-metastatic disease. As illustrated in Appendix Section H, Appendix Figure 14, there is little difference in effect sizes across the three subgroups that could indicate lower-value utilization. To rule out the possibility that all marginal users are only filling once, which could be an indicator of a low-value prescription that was poorly tolerated, the effect of physician-pharmacy integration on second-time fills was also evaluated. Results reveal similar effects on second-time fills as first-time fills, suggesting marginal users are not only filling once (Appendix Section G, Appendix Table A6). I then sought to derive an estimate for the cost-effectiveness for high-cost oral cancer drugs, assuming marginal use cases are similar to those evaluated in literature. Specifically, a literature review of 25 cost-effectiveness studies for the top 11 high-cost oral cancer treatments, which account for 80% of spending during the study period, was performed; see Appendix Section H for details on studies included. This literature review included all U.S. based cost-effectiveness studies that provide an estimate for the incremental cost effectiveness per QALY for a regimen with a high-cost oral cancer treatment relative to a regimen without a high-cost oral cancer treatment, identified via the Tufts Cost Effectiveness Registry. Fig. 7 reports the distribution of estimates for each drug for which at least one estimate was available; variation across studies within drug is related to clinical factors such as stage and bio-markers, alternative treatment considered, time period, and methods used. The weighted average cost-effectiveness across drugs using the median estimate per drug, and weighting across drugs by share of Medicare Part D spending during the study period, yields an estimated average cost-effectiveness of $167K per QALY. Range of literature-based estimates for incremental cost per quality-adjusted life year of high-cost oral cancer treatments. Note: This analysis considers all U.S.-based cost effectiveness studies found in the Tufts Cost Effectiveness Analysis Registry containing reference to the following 11 drugs: lenalidomide, imatinib mesylate, ibrutinib, abiraterone acetate, palbociclib, enzalutamide, ruxolitinib phosphate, pomalidomide, erlotinib, dasatinib, and nilotinib. The final sample includes 25 studies that provide an incremental cost per quality-adjusted life year (QALY) for at least 1 regimen containing a high-cost oral cancer treatment vs. regimen without a high-cost oral cancer treatment (or placebo). This excludes studies where all regimens include high-cost oral drugs, the high-cost oral drug is dominated or dominant relative to an alternative, the high-cost drug produces lower health benefits relative to the alternative, or no cost-effectiveness analysis of a treatment regimen was done. For each study-drug combination, I collect any estimates for incremental cost per QALY for each drug in sample; if the same study reports multiple estimates for a single drug, I take the midpoint. 1 study reported estimates for 2 drugs, resulting in a final sample of 26 estimates spanning 7 drugs. The minimum, 25 percentile, 50 percentile, 75 percentile, and maximum average cost per QALY is estimated using spend weighted average of each drugs’ minimum, 25 percentile, 50 percentile, 75 percentile, and maximum estimate respectively. The share of spending accounted for by each molecule is calculated based on all Part D spending from 2006 to 2019. However, there are several challenges with using this estimate to evaluate welfare. First, if marginal uses are not those evaluated in the literature, the true cost per QALY could deviate meaningfully. Indeed, the literature review suggests that a range for the weighted average from $65K to $1.3M per QALY is plausible, and even using the 25th or 75th percentile estimate for each drug yields a range of $90K to $266K per QALY (Fig. 7). Second, the cost-effectiveness estimates consider the point-of-sale drug price. This perspective may be sensible for public payers allocating limited funds to the highest value uses. However, another perspective could consider only the marginal costs of production and fiscal externalities of raising public funds. This perspective may be more sensible if the public payer can raise funds and would imply a far more favorable cost-effectiveness. Finally, there is considerable disagreement regarding societal willingness-to-pay for a QALY. From the 1990s-2010s, $50K per QALY was the most cited threshold. However, the World Health Organization, suggests a cost-effectiveness threshold of 2–3 times per capita income, which would imply a U.S. threshold of $110K - $160K per QALY (Neumann et al., 2014). Meanwhile, a review of studies evaluating willingness-to-pay per QALY, suggests a median estimate of $265K per QALY (Hirth et al., 2000). This is consistent with others who have argued for a threshold $200K - $300K per QALY, using the ratio of health gains to health care costs over time (Braithwaite et al., 2008). Nonetheless, the cost-effectiveness of marginal uses of high-cost oral cancer treatments and the appropriate threshold for cost-effectiveness will dominate the welfare analysis. This is because, as shown in Appendix Section H, Appendix Table 13, the plausible welfare effects from the secondary drivers are smaller than the variance implied by the range of plausible cost effectiveness estimates and thresholds. This is true irrespective of whether the relevant measure of utilization is the number of patients initiating treatment or total drug-days supplied (Appendix Section H, Appendix Table 14). Thus, a better empirical understanding of the marginal use cases is an important area for future research. Conclusion Integration between physicians and pharmacies is becoming increasingly common for self-administered specialty drugs and is likely to continue. The creation of the Medicare Part D in 2006 has spurred innovation in oral drugs (Blume-Kohout and Sood, 2013), with oral cancer treatments being among the fastest growing segments (Kyle et al., 2022). Increasing practice consolidation is also likely to increase the share of physicians employed at organizations with pharmacies (Nikpay et al., 2018). Furthermore, the use of in-house pharmacies has also grown in other specialty segments, suggesting this phenomenon extends beyond oncology. Finally, with efforts by national pharmacy chains such as CVS, Walgreens, and Amazon to enter primary care, considerations regarding integration between physicians and pharmacies may eventually expand more meaningfully beyond specialty medicine. This study finds that physician–pharmacy integration increases the likelihood and speed of treatment initiation, likely due to operational efficiencies, without anti-competitive effects. Prices at in-house pharmacies are 1.0% lower than at external pharmacies suggesting limited anti-competitive and potential pro-competitive effects. Physician–pharmacy integration also increases the share of patients with first-time fills for high-cost oral cancer treatments by 6.2% and reduces the time-to-fill when drugs under prior authorization. This suggests physician practices with in-house pharmacies may be able to accomplish prior authorization faster, which smooths access to medication. However, as in limited prior work, I do not find evidence of substitution to higher cost regimens or improved adherence following treatment initiation (Kanter et al., 2024). This may be in part because, in this clinical scenario, the choice to prescribe more expensive or longer duration oral cancer treatments may not discretionary for most patients (Chandra and Skinner, 2012). This study is among the first systematic studies of physician–pharmacy integration for self-administered drugs in the US, and represents the largest study of physician–pharmacy integration to date. Prior work has been limited to one study of commercially insured patients treated by independent oncology practices in select states, which generally found no statistically significant impacts of physician–pharmacy integration in oncology on likelihood of using oral chemotherapy, cost of care, medication adherence, or days between diagnosis and treatment initiation (Kanter et al., 2024). However, relative to prior work this study was likely more equipped to detect effects due to the substantially larger, national population of cancer patients, inclusion of independent and hospital-affiliated practices, ability to exclude integration events where the in-house pharmacy was not meaningfully used, and evaluation of additional outcomes that may be impacted by physician–pharmacy integration (e.g., outcomes related to high-cost medicines, the time between a prescription and fill, and prices). Perhaps as a result, our findings related to increased and faster treatment initiation of oral cancer treatments differ from (Kanter et al., 2024). These findings also stand in contrast to much of the existing literature on vertical integration in US health care markets more broadly, especially related to physician-hospital integration, which generally finds price increases, spending increases, and limited quality improvements related to operational efficiencies (Post et al., 2018; Machta et al., 2019; Amado et al., 2022). This context therefore highlights three key facts important for vertical integration policy. First, vertical integration among healthcare providers does not necessitate price increases, even when medical and pharmacy services are jointly bargained over. This may be especially so when vertical integration reflects entry of a competitor with limited bargaining power. Second, organizational financial incentives resulting from vertical integration, such as those created by the 340B program, may not necessarily impact utilization, particularly when substitution opportunities are not therapeutically equivalent. Finally, vertical integration can, indeed, result in operational efficiencies improving some dimensions of quality — in this case the timeliness of treatment initiation. These may be particularly salient during care transitions requiring coordination from multiple entities, such as obtaining prior authorization and beginning a high-cost oral cancer treatment. Nonetheless, this study has limitations that suggest areas for future research. First, it is unclear whether the increase in treatment initiation is welfare-improving as the cost-effectiveness of treatment initiation varies considerably by unobserved clinical characteristics (e.g., staging and biomarkers). This suggests value in analyzing the impact of integration in data with richer patient characteristics, to better understand the marginal population. Second, while this study evaluates the overall impact of physician–pharmacy integration on a practices’ patients, it does not disentangle the impact of integration on patients who use in-house versus external pharmacies. This is challenging to do credibly in this context as in-house pharmacies are generally included in Part D networks and thus there is potential of unobserved selection into use of in-house pharmacies. However, future work may consider evaluating the impact of in-house pharmacy launch on patients for whom the in-house pharmacy is explicitly out-of-network to understand spillover effects on this population. Third, this study has limited power to evaluate certain aspects of heterogeneity, such as across practice characteristics that are often co-linear. This suggests value to replicating analyses in larger samples. Finally, this analysis was limited to physician–pharmacy integration in oncology. Physician–pharmacy integration has also grown in other specialties suggesting a need for broader evaluations (Kakani et al., 2024). Supplementary Material Declaration of competing interest Dr. Kakani gratefully acknowledges support from the National Institute on Aging T32-AG000186 for this work. Unrelated to this work, Dr. Kakani acknowledges research support from the Commonwealth Fund, Arnold Ventures, the National Institute for Health Care Management, the National Cancer Institute, the Centers for Disease Control. Dr. Kakani has no other relevant conflicts of interest to declare. I thank David Cutler, Nancy Keating, Amitabh Chandra, Meredith Rosenthal, Mark Shepard, Timothy Layton, Mary Beth Landrum, Stacie Dusetzina, Aaron Mitchell, Mariana Socal, Genevieve Kanter, and conference participants at the American Society of Clinical Oncology Quality Care Symposium and the American Society of Health Economists Annual Meeting for their insightful comments. I also thank Mohan Ramanujan, Maurice Dalton, Brian Young, Nancy Beaulieu, Zakaria El Amrani El Idrissi, Bernard Black, Timea Viragh, Ali Mogtaderi, Sunita Desai, and Michael McWilliams for assistance with data. I gratefully acknowledge support from the National Institute on Aging T32-AG000186 for this work. The content is solely the responsibility of the author. This research was approved by the Institutional Review Boards of Harvard University and the National Bureau of Economic Research. Supplementary data Supplementary material related to this article can be found online at https://doi.org/10.1016/j.jhealeco.2025.103085. Footnotes See Gal-Or (1999), Peters (2014) for theoretical foundations and Ciliberto and Dranove (2006), Baker et al. (2014), Neprash et al. (2015), Carlin et al. (2015), Scheffler et al. (2018), Capps et al. (2018), Lin et al. (2021), Curto et al. (2022) for empirical evidence. See Hart et al. (1990), Ordover et al. (1990), Rasmusen et al. (1991) for theoretical foundations and Cutler et al. (2020), Richards et al. (2022) for empirical evidence. Pharmacy DIR fees are the term used in Medicare Part D, but commercial plans can also have similar fees. CMS Final Rule 87 FR 27704 made retroactive pharmacy DIR fees illegal in Medicare Part D beginning in 2024. These estimates do not consider pharmacy DIR fees, which represented less than 0.2% of gross sales in Medicare Part D in 2013 but may have grown to 2% of gross sales by 2019 (Fein, 2020). Nonetheless, re-analysis of these data suggest retail pharmacies’ margins net of DIR fees likely continue to be 3%–4% on average and are similar for high-cost oral cancer treatments on average (Appendix Section B, Appendix Table A1). Humana Pharmacy Solutions contracts with SS&C Health for pharmacy claims administration. Thus, certain estimates of PBM market share in Medicare will list SS&C Health rather than Humana Pharmacy Solutions. Select oral drugs with infused equivalents, are covered by Part B. However, Medicare Part D accounted for > 98% of spending on oral cancer treatments in 2017 among Medicare beneficiaries enrolled in Fee-for-Service (Parts A and B) and Part D (author’s analysis). Medicare Part D may also cover infused chemotherapies for patients receiving such treatments at home or through their prescription benefit; however these use cases are relatively uncommon in Medicare (Anderson et al., 2021) Beginning in 2025, government, plan, manufacturer, and patient liabilities have changed to be 20%, 60%, 20%, and 0% respectively. State or federal policy variation is not leveraged in the empirical analysis for multiple reasons. First, policy restrictions only apply to independent, physician-owned practices and their pharmacies. Second, only two states (Massachusetts, Montana) imposed significant limitations on both physician dispensing and physician ownership of retail pharmacies throughout the study period. Finally, virtually all sizable policy changes have been too recent (2018 onwards) to facilitate study. See Ciliberto and Dranove (2006), Baker et al. (2014), Neprash et al. (2015), Carlin et al. (2015), Scheffler et al. (2018), Capps et al. (2018), Lin et al. (2021), Curto et al. (2022). These include office visits, emergency room visits, inpatient consult claims, etc. Appendix Section C.2 offers details on identifying these claims. Practices that only use in-house pharmacies for oral cancer treatments very rarely, and thus are not classified as treated practices, may reflect multiple scenarios including (a) practices part of multi-specialty organizations with retail pharmacies that only occasionally fill oncology prescriptions (b) practices that only opportunistically dispense samples when available (e.g., when provided by marketers), and (c) practices with in-house pharmacies only available at a very limited number of practice sites or with very limited stock. None of the independent practices in the event sample were located in Massachusetts or Montana, consistent with state policies restricting physician ownership of in-house pharmacies and physician dispensing in these states. All practices are observed for three years before and after the event, but some practices are unobserved 4–5 years before or after the event due to closure, acquisition, or study period constraints. The main specification includes five years before and after the event to allow ample time to evaluate pre-period trends and changes associated with the event. Nonetheless, I evaluate the sensitivity of results to inclusion of only the balanced sample before and after the event. If a patients’ first time filling lenalidomide (a high-cost oral cancer treatment) in Medicare Part D occurs in 2013 and the patients’ first time filling imatinib mesylate (a high-cost oral cancer treamtent) occurs in 2016, then he is said to have a first-time fill in 2013 and 2016. For 6% of first fills, there is no consult with the prescribing oncologist in the six months prior, which may reflect limited cases where the prescription occurs outside of an office visit context. The median number of days between a first-fill and the last physician consult is 2 days, but is lower for low-cost drugs that may be more easily accessible (0 days) than for high-cost drugs (6 days), suggesting face-validity of the measure. The numerator is the total days supplied of the drug between the first and last fill within 12 months. The denominator is the total days between the first and last fill + the days supplied at the last fill. Financial assistance programs are generally offered by external charitable organizations, often funded by manufacturers, rather than by the practice or pharmacy (Dafny et al., 2022). This is based on the median number of oncologists in the treatment group (7). Thus above median practices are those with 8+ oncologists. This fixed effect is not interacted with an indicator for above median practice size because the experiment definition, which includes cells of practice size, is colinear with whether the practice is above median practice size. While it is possible that side effects of these drugs may make them net harmful for some patients, this is unlikely in part because side-effect profiles of oral therapies are generally “somewhat more favorable” than infused therapies (Halfdanarson and Jatoi, 2010). Whether a medication is tolerable for a specific patient is also likely highly uncertain prior to the patient using the drug at least once, as patients respond differently to treatment. If the in-house pharmacy does not offer delivery of medication by mail, it is also possible that integration may reduce patient welfare if patients now need to travel to the practice in-person to receive medications. However, we do not consider this in our analysis as many in-house pharmacies do offer delivery of medicines by mail, though the frequency of mail delivery is unobserved in the data. Moreover, some patients may prefer an in-person to a mail-order option. References Associated Data Supplementary Materials