Real world comparative outcomes of adjuvant carboplatin versus cisplatin-based chemotherapy in stage IB–IIIA non-small cell lung cancer: a propensity-weighted analysis Authors: Molly Tokaz, Brendan T. Heiden, Daniel B. Eaton, Shadia Jalal, Nahom Seyoum, Whitney S. Brandt, Nikki E. Rossetti, Su-Hsin Chang, Yan Yan, Varun Puri, Martin W. Schoen, Theodore S. Thomas Published: 2026 ABSTRACT BackgroundPlatinum-doublet adjuvant chemotherapy is standard of care for resected non-small cell lung cancer (NSCLC). Cisplatin has been historically favored for modest survival benefits, though its substantial toxicity often limits treatment completion in routine practice. Carboplatin is better tolerated and frequently substituted; however, comparative real-world effectiveness and safety remain unclear. We evaluated outcomes of cisplatin- versus carboplatin-based adjuvant therapy within the Veterans Health Administration (VHA).MethodsWe conducted a retrospective cohort study of Veterans with resected stage IB–IIIA NSCLC treated with adjuvant cisplatin- or carboplatin-based doublets from 2006–2020 using national VHA data. Inverse probability of treatment weighting (IPTW) balanced demographic, clinical, and surgical characteristics. Outcomes included overall survival (OS), recurrence-free survival (RFS), renal toxicity, hospitalization, adverse event rate and treatment completion.ResultsAmong 465 Veterans, 242 received cisplatin and 223 carboplatin. Carboplatin recipients were older with more comorbidities. After IPTW adjustment, cisplatin was associated with higher renal toxicity (≥30% eGFR decline: 42.7% vs. 21.3%, P<0.001), greater overall adverse event burden (3.41 vs. 2.75, P=0.002), higher rates of hospitalization (46.8% vs. 35.2%, P=0.02) and lower treatment completion (37.7% vs. 55.5%, P<0.001). Five-year OS and RFS numerically favored cisplatin (61.0% vs. 48.8%, P=0.17; 63.9% vs. 56.6%, P=0.12).ConclusionsCisplatin-based adjuvant chemotherapy was associated with higher OS and RFS though with significantly greater toxicity, hospitalization, and lower treatment completion. These findings highlight the importance of considering the tradeoff of tolerability and treatment completion with efficacy in real-world populations to allow individualized regimen selection, particularly when cisplatin is not feasible. Highlight box Key findings • In this large retrospective Veterans Health Administration cohort, cisplatin-based adjuvant chemotherapy was associated with significantly greater renal toxicity, overall adverse event burden, hospitalization rates, and lower treatment completion compared with carboplatin-based therapy. • Cisplatin demonstrated numerically improved overall survival and recurrence-free survival; these differences were not statistically significant after inverse probability of treatment weighting (IPTW) adjustment but remained clinically significant. What is known and what is new? • Cisplatin has historically been considered the preferred platinum backbone for adjuvant treatment of resected non-small cell lung cancer (NSCLC) based on modest survival benefits demonstrated in randomized trials, though its use is frequently limited by toxicity. Comparative real-world toxicity and treatment completion data evaluating cisplatin versus carboplatin in older and medically complex populations remain limited. • This study provides one of the most comprehensive real-world comparisons of cisplatin- versus carboplatin-based adjuvant chemotherapy within a nationally integrated veteran population, incorporating detailed longitudinal toxicity, hospitalization, and treatment completion outcomes using IPTW methodology. What is the implication, and what should change now? • These findings support a more individualized approach to platinum selection in resected NSCLC, particularly for older or medically frail patients at increased risk for treatment-related toxicity and treatment discontinuation. • In contemporary practice, where completion of multimodality perioperative therapy with targeted therapies is increasingly important, clinicians should carefully weigh the modest potential efficacy advantage of cisplatin against its substantially greater toxicity burden and effects on treatment completion. Introduction Early-stage localized non-small cell lung cancer (NSCLC) constitutes approximately 23% of NSCLC diagnoses in the United States (1). Treatment is with curative intent, with an approximately 65% 5-year overall survival (OS) rate (1). Surgical resection followed by adjuvant cytotoxic chemotherapy with a platinum doublet—cisplatin or carboplatin plus an additional chemotherapy agent—is the standard of care for patients with select stage IB and above NSCLC who did not receive perioperative chemoimmunotherapy (2-4). These regimens have been used for years; however, some individuals with resected lung cancer are not cisplatin candidates due to impairments in baseline organ function and comorbid conditions. Furthermore, some patients are not willing to accept the higher risk of toxicity for modest survival gains conferred by cisplatin use. The initial landmark studies establishing adjuvant platinum therapy demonstrated modest survival benefits with cisplatin, cementing its role as the standard chemotherapy backbone. The LACE analysis, a pooled retrospective study, along with the ANITA and IALT randomized trials demonstrated 5-year OS benefits ranging from 4–8.6% (5-7). Cisplatin has never been directly compared to carboplatin in a large, randomized fashion, so much of the understanding of carboplatin efficacy is based on indirect comparisons or retrospective analyses. In 2008, the randomized CALGB 9633 studied carboplatin versus observation in stage I patients and did not demonstrate a significant OS benefit, calling into question carboplatin efficacy (8). Since that time, additional analyses of higher stage patients have demonstrated survival benefit with carboplatin (9) and further retrospective studies and meta-analyses attempting to compare cisplatin and carboplatin-based regimens have failed to demonstrate a significant survival difference between the agents (10-16). Taken as a whole, cisplatin remains the standard of care for adjuvant treatment; however, carboplatin is widely considered a reasonable alternative in resected patients not fit for the gold-standard cisplatin (13,15-18) . The use and duration of the platinum-doublets in real-world populations is often limited by toxicity. Cisplatin is associated with an increased risk of emetogenicity, ototoxicity, peripheral neuropathy, and nephrotoxicity which can lead to increased risks of hospitalization and treatment discontinuation, particularly in those with increased medical comorbidities and/or elderly (16,19,20). In particular, platinum doublets are associated with a 43% increase in the risk of platinum-induced nephrotoxicity in older individuals, with severe and dose limiting toxicities due to cisplatin use (21). Carboplatin-based doublets studied in elderly and comorbid populations continue to show survival benefit (22-24), and are widely used in practice for individuals deemed unfit for cisplatin and the elderly (>70 years), including in 65% of veterans (10), based on a prior study. Contemporary clinical trials in perioperative and adjuvant settings have variably permitted carboplatin-based regimens, reflecting increasing adoption of carboplatin into adjuvant chemotherapy regimens for resectable NSCLC (25-30). Still, the comparative effectiveness and tolerability of carboplatin- versus cisplatin-based adjuvant chemotherapy in older, comorbid populations remain poorly characterized, particularly in integrated health systems such as the Veterans Health Administration (VHA). The VHA provides a unique opportunity to study comparative treatment effectiveness in a real-world population. Given the long-standing availability of both cisplatin- and carboplatin-based adjuvant regimens and the large, nationally integrated VHA electronic health record system, an evaluation of toxicity, treatment completion and survival outcomes associated with each agent in this setting is both feasible and necessary. We conducted a retrospective cohort study using a uniquely curated VHA dataset that includes all Veterans with resected NSCLC from 2006 to 2020. We compared the real-world effectiveness, toxicity, and regimen completion of adjuvant cisplatin- versus carboplatin-based doublets among Veterans with resected stage IB–IIIA NSCLC. We aim to address the real-world clinical trade-offs between efficacy and tolerability, with the goal of informing clinician decision making for optimal adjuvant treatment selection for individuals with resected NSCLC. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0658/rc). Methods This retrospective cohort study included Veterans with clinical stage I NSCLC who underwent pulmonary resection at a national VHA facility between Oct 1st, 2006 and December 31st, 2020. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The VA St. Louis Health Care System Research and Development Committee (#1214632, approved 08/02/2019) the study protocol, the Institutional Review Board (IRB) waived the requirement for informed consent due to the retrospective and deidentified nature of the analysis. The study used a unique and highly curated dataset that we have described several times previously (31). This dataset was developed through an intensive, multi-year effort by a team of analysts by conducting manual medical record reviews—enhanced by natural language processing techniques—to extract exhaustive patient-, tumor-, and treatment-related covariates. This process encompassed all Veterans with resected NSCLC from 2006 to the present, resulting in an unparalleled dataset that reflects both clinical nuance and real-world complexity. Patient data were extracted from the VA Informatics and Computing Infrastructure and Corporate Data Warehouse (CDW), which integrate multiple national clinical and administrative datasets including the VA Surgical Quality Improvement Program (VASQIP) and CDW-Oncology Raw Domain (32). International Classification of Diseases (ICD) for Oncology codes were used to identify NSCLC diagnoses; procedures were identified by a combination of ICD codes and Current Procedural Technology Codes (CPT). The selected cohort included Veterans 18 years and older with presurgical tumors ≤5 cm and no lymph node involvement (Clinical stage I based on the American Joint Committee on Cancer’s 7th edition), and who received curative-intent lung resection followed by either adjuvant Carboplatin or Cisplatin. Veterans were excluded if they had delayed adjuvant therapy (≥180 days from surgery completion), underwent more than 4 cycles of adjuvant chemotherapy, had an R1 section, were found to be stage 1A (i.e., ineligible for adjuvant chemotherapy based on contemporary guidelines) or IV disease. Defining treatment of cisplatin or carboplatin We used the VA Corporate Data Warehouse (CDW) for tracking IV medicine administration starting from day after operation up until 1 year after the operation for drug level information inside the VA. For Veterans who treated outside the VA, the Fee Basis Claim system or the IVC, CPT Codes (J9060, J9065, J9045), which identify community care paid for by the VA, as a claims level data point. These tables were mutually exclusive. All drug administrations from the respective data tables were included up through 1-year post-surgery to allow for exclusion of ineligible Veterans. Veterans were allocated by the first drug received to account for Veterans who switched regimens. Veterans are classified by the initial platinum agent received, and if therapy was changed after initial treatment this switch was captured as a data point. Covariates Covariates utilized in the analysis included sex, race, and age, body mass index (BMI), smoking status at time of surgery, comorbidities, distance to clinic and others. Comorbidities were calculated with a composite Charlson-Deyo score using ICD coded comorbidities from 5 years prior to 1 month after surgery (33). Smoking status was defined using our previously defined algorithms (34,35). Distance to the VHA clinics were calculated from the center of the Veterans’ residential zip codes to the treatment facility location. Oncologic surgery-related covariates included tumor factors (size, grade, histology), year of operation, hospital case load (defined as the volume of lung cancer cases treated in a specific facility in the year preceding the surgery), surgical approach [video-assisted thoracoscopic surgery (VATS) or thoracotomy], type of operation (lobectomy, segmentectomy, wedge resection, or pneumonectomy), and final pathologic stage (32). Outcomes Efficacy outcomes were OS and recurrence-free survival (RFS). OS was assessed using VA death ascertainment file and was defined as the time from treatment initiation to all-cause mortality (with censoring date being January 15, 2025). RFS was defined as time from treatment initiation to recurrence (with death treated as a competing event), as previously described by our group (33). Toxicity outcomes included the incidence of treatment-related adverse events, particularly ototoxicity, nephrotoxicity, and peripheral neuropathy. Ototoxicity was defined as Veterans having an ICD diagnosis or new hearing aid testing (defined as not having a hearing aid prior to treatment initiation up until 270 days after) or having visit with audiology within 270 days of treatment initiation. Nephrotoxicity was assessed in several ways including acute kidney injury (AKI) by ICD, AKI by 30% decrease in eGFR, minimum eGFR post-treatment start, and percent change in baseline estimated glomerular filtration rate (eGFR) values. eGFR was calculated using blood chemistry tests and was defined using LOIN-C (Logical Observation Identifiers, Names and Codes) codes. Baseline eGFR levels were assessed within 1 year to 14 days prior to resection. Maximum post-treatment eGFR changes were assessed after the date of treatment initiation (up to 1 year after resection). Peripheral neuropathy was defined as Veterans who received a new prescription for gabapentin or pregabalin within 270 days of treatment initiation. Other secondary endpoints included emesis, dehydration, neutropenia, and thrombocytopenia, hospitalization, emergency department visits, switching agents, and composite outcomes of multiple adverse events (e.g., any adverse outcome of interest, number of adverse events of interest, etc.). Inverse probability treatment weighting (IPTW) Due to differences in baseline characteristics between carboplatin and cisplatin, IPTW was employed to create balanced cohorts. For IPTW, we weighted the cohort towards the carboplatin group for comparing treatment-related outcomes. For long-term outcomes, Veterans who switched therapy (i.e., carboplatin to cisplatin, or vice versa) were removed and then IPTW was re-matched to ensure the cohorts remained balanced. IPTW balancing was based on clinically relevant covariates including age, sex, race, CCI, BMI, smoking status, distance from hospital, tumor histology, eGFR, and surgical year (36). Differences between weighted groups were displayed with standardized mean differences (SMD), with a goal of SMD of less than 0.10 based on prior literature. Statistical analysis Descriptive statistics for the cohort utilized means (standard deviations) for all continuous variables and frequencies (proportions) for all categorical variables. OS was assessed using multivariable Cox proportional hazards model and displayed using the Kaplan-Meier method. RFS utilized a multivariable competing risk model (Fine and Gray sub-distribution hazard function) with recurrence as outcome and death as a competing event. There were minimal missing data points in this data set; when present, they were displayed using unknown categories. All statistical tests were two-sided. P values less than 0.05 were considered statistically significant. All analyses were performed in SAS version 9.3 (SAS Institute Inc., Cary, NC, USA) and R Studio (version 1.3). Results Baseline cohort characteristics: A total of 12,271 Veterans with resected clinical stage I NSCLC were considered for the study, of which 675 Veterans had completely resected pathological stage IB–III NSCLC and received adjuvant chemotherapy. From this stage selected population, the cohort was further refined to exclude patients with missing demographic variables or with early recurrence, resulting in a final analytic cohort of 465 veterans, including 242 treated with cisplatin and 223 treated with carboplatin-based chemotherapy (Figure 1). Carboplatin recipients were older (66.1 vs. 64.5 years; P=0.01) with higher comorbidities (CCI 2.67 vs. 2.15; P=0.002), while sex, race, rurality and smoking status were comparable between groups. Cisplatin Veterans had higher baseline renal function (eGFR 87 vs. 81 mL/min, P<0.001), and BMI distributions differed modestly (P=0.006) (Table 1). CONSORT diagram. IPTW, inverse probability of treatment weighting. Table 1Demographics and baseline disease characteristics of carboplatin and cisplatin cohorts, pre-IPTW weighted (total n=465)VariableCarboplatin (n=223)Cisplatin (n=242)P valueSMDAge (years)66.18±7.6964.49±6.760.010.234CCI2.67±1.902.15±1.800.0020.283Sex0.480.086 Female8 (3.6)13 (5.6) Male215 (96.4)229 (94.6)Race0.510.106 Black44 (19.7)38 (15.7) Other2 (0.9)2 (0.8) White177 (79.4)202 (83.5)Smoking status0.620.090 Current130 (58.3)150 (62.0) Former85 (38.1)86 (35.5) Never8 (3.6)6 (2.5)BMI (kg/m2)0.0060.358 <18.53 (1.3)4 (1.7) 18.5–24.964 (28.7)74 (30.6) 25–29.973 (32.7)89 (36.8) 30–34.967 (30.0)41 (16.9) 35+16 (7.2)34 (14.0)Distance to clinic (rurality)0.970.019 ≤1039 (17.5)42 (17.4) 11–5091 (40.8)101 (41.7) 50+93 (41.7)99 (40.9)Hospital dx caseload111.51±55.64111.23±49.850.950.005Histology0.360.132 Adenocarcinoma116 (52.0)140 (57.9) Squamous cell carcinoma70 (31.4)71 (29.3) Other37 (16.6)31 (12.8)Tumor size (mm)0.440.180 ≤107 (3.1)14 (5.8) 11–2052 (23.3)54 (22.3) 21–3041 (18.4)55 (22.7) 31–4074 (33.2)72 (29.8) 41–5049 (22.0)47 (19.4)Surgical delay0.200.127 No158 (70.9)185 (76.4) Yes65 (29.1)57 (23.6)Type of resection0.650.117 Lobectomy174 (78.0)198 (81.8) Pneumonectomy10 (4.5)11 (4.5) Segmentectomy13 (5.8)13 (5.4) Wedge resection26 (11.7)20 (8.3)Incision0.030.207 Thoracotomy125 (56.1)160 (66.1) VATS98 (43.9)82 (33.9)Lymph node count0.090.234 08 (3.6)4 (1.7) 1–447 (21.1)35 (14.5) 5–970 (31.4)76 (31.4) 10+98 (43.9)127 (52.5)Baseline albumin (g/dL)3.9 [3.7, 4.2]3.9 [3.7, 4.2]0.960.033Baseline eGFR (mL/min/1.73 m2)81 [63, 96]87 [74, 98]<0.0010.366Baseline hemoglobin (g/dL)14.3 [13.2, 15.2]14.2 [13.3, 15.0]0.700.064Baseline platelets (K/μL)225 [186, 263]233 [194, 280]0.120.138Pathologic stage0.020.248 I76 (34.1)56 (23.1) II91 (40.8)120 (49.6) III56 (25.1)66 (27.3)Data are presented as mean ± standard deviation, n (%) or median [interquartile range]. BMI, body mass index; CCI, Charlson comorbidity index; dx, diagnosis; eGFR, estimated glomerular filtration rate; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference; VATS, video-assisted thoracoscopic surgery. Demographics and baseline disease characteristics of carboplatin and cisplatin cohorts, pre-IPTW weighted (total n=465) Data are presented as mean ± standard deviation, n (%) or median [interquartile range]. BMI, body mass index; CCI, Charlson comorbidity index; dx, diagnosis; eGFR, estimated glomerular filtration rate; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference; VATS, video-assisted thoracoscopic surgery. Surgical factors also varied in several ways, including that cisplatin Veterans more often underwent thoracotomy (66.1% vs. 56.1%) whereas carboplatin Veterans more frequently had minimally invasive resections (P=0.03), though extent of surgery and lymph node sampling did not significantly differ between the cohorts. Rate of surgical delay did not differ significantly between the groups (29.1% for carboplatin vs. 23.6%, P=0.20) (Table 1). Histology distribution was broadly balanced, as were pretreatment biochemical baselines including albumin, hemoglobin and platelet count. Carboplatin patients were associated with a higher pathologic stage I (34.1% vs. 23.1%) and lower pathologic stage III (25.1% vs. 27.3%); P=0.02 (Table 1). IPTW cohort characteristics The cohort of 465 veterans were then matched using IPTW weighting resulting in effective populations of 465 veterans in the carboplatin evaluable group and 459 in the cisplatin evaluable group (Table 2). After applying the clinical weighting criteria, no significant differences remained between the cohorts. Specifically, there was no significant difference in pathologic staging for carboplatin vs. cisplatin: pathologic stage I (29.9% vs. 29.6%) and pathologic stage III (24.5 vs. 25.6) P=0.97; in the matched cohorts. Additionally, after propensity matching, no significant difference remained in age, CCI, BMI or surgical factors (the standardized mean differences for IPTW covariate balance is available in Figure S1 for reference). Table 2Demographics and baseline disease characteristics of carboplatin and cisplatin cohorts, IPTW weightedVariableCarboplatin (n=465.7)Cisplatin (n=459.4)P valueSMDPropensity score0.48±0.190.47±0.180.780.031Age (years)64.99±7.9265.15±6.760.830.022CCI2.39±1.822.37±1.900.950.007Sex0.910.011 Female23.2 (5.0)21.7 (4.7) Male442.5 (95.0)437.7 (95.3)Race0.960.026 Black79.4 (17.1)74.1 (16.1) Other3.9 (0.8)3.7 (0.8) White382.3 (82.1)381.7 (83.1)Smoking status0.970.023 Current284.1 (61.0)276.9 (60.3) Former167.7 (36.0)169.9 (37.0) Never13.8 (3.0)12.6 (2.7)BMI (kg/m2)0.990.052 <18.56.7 (1.4)7.5 (1.6) 18.5–24.9142.4 (30.6)139.4 (30.3) 25–29.9153.4 (32.9)160.7 (35.0) 30–34.9108.7 (23.3)100.8 (21.9) 35+54.5 (11.7)51.0 (11.1)Distance to clinic (rurality)0.970.024 ≤1076.8 (16.5)78.9 (17.2) 11–50191.4 (41.1)190.7 (41.5) 50+197.4 (42.4)189.8 (41.3)Hospital dx caseload112.70±54.10109.94±49.090.580.053Histology0.990.010 Adenocarcinoma254.2 (54.6)250.6 (54.5) Squamous cell carcinoma144.8 (31.1)141.6 (30.8) Other66.7 (14.3)67.2 (14.6)Tumor size (mm)0.990.035 ≤1019.1 (4.1)21.0 (4.6) 11–20111.1 (23.9)105.4 (23.0) 21–3089.8 (19.3)91.3 (19.9) 31–40145.4 (31.2)144.8 (31.5) 41–50100.4 (21.6)96.9 (21.1)Surgical delay0.770.028 No344.7 (74.0)345.6 (75.2) Yes121.0 (26.0)113.8 (24.8)Resection0.990.022 Lobectomy372.4 (80.0)369.8 (80.5) Pneumonectomy22.1 (4.7)20.5 (4.5) Segmentectomy25.1 (5.4)25.8 (5.6) Wedge resection46.0 (9.9)43.3 (9.4)Incision0.730.035 Thoracotomy280.0 (60.1)284.1 (61.8) VATS185.7 (39.9)175.3 (38.2)Lymph node count0.990.020 011.6 (2.5)12.1 (2.6) 1–481.5 (17.5)77.7 (16.9) 5–9151.1 (32.5)151.9 (33.1) 10+221.4 (47.5)217.8 (47.4)Baseline albumin (g/dL)3.9 [3.7, 4.1]4.0 [3.7, 4.2]0.440.016Baseline eGFR (mL/min/1.73 m2)88 [68, 97]82 [69, 97]0.060.005Baseline hemoglobin (g/dL)14.3 [13.2, 15.1]14.3 [13.3, 15.0]0.880.014Baseline platelets (K/μL)230 [188, 278]231 [193, 275]0.750.008Pathologic stage0.970.024 I139.4 (29.9)136.2 (29.6) II212.1 (45.5)205.8 (44.8) III114.2 (24.5)117.5 (25.6)Data are presented as mean ± standard deviation, n (%) or median [interquartile range]. BMI, body mass index; CCI, Charlson comorbidity index; dx, diagnosis; eGFR, estimated glomerular filtration rate; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference; VATS, video-assisted thoracoscopic surgery. Demographics and baseline disease characteristics of carboplatin and cisplatin cohorts, IPTW weighted Data are presented as mean ± standard deviation, n (%) or median [interquartile range]. BMI, body mass index; CCI, Charlson comorbidity index; dx, diagnosis; eGFR, estimated glomerular filtration rate; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference; VATS, video-assisted thoracoscopic surgery. Safety and tolerability In the unmatched cohort (Table 3), cisplatin was associated with a higher incidence of renal toxicity, with acute kidney injury defined by ≥30% eGFR decline occurring in 44.2% of Veterans compared with 20.6% in the carboplatin group (P<0.001). Cisplatin was also associated with greater rates of emesis (18.6% vs. 11.2%, P=0.03). Overall symptom-based adverse events were more frequent with cisplatin (86.0% vs. 74.9%, P=0.004), and cisplatin Veterans were more likely to require hospitalization (46.3% vs. 35.4%, P=0.02), with higher mean number of hospitalization events (1.69 vs. 1.29, P=0.033). Emergency department visits were more common among Veterans who received cisplatin (48.8% vs. 43.0%, P=0.25), though did not reach significance. The mean number of adverse events was higher with cisplatin (3.41 vs. 2.76, P=0.001). Pre-weighting Treatment completion differed significantly, with only 38.0% of cisplatin Veterans completing four cycles without switching compared to 56.1% of carboplatin Veterans (P<0.001). Table 3Table of safety and tolerability for carboplatin vs. cisplatin adjuvant therapy for unmatched cohort (including switchers)OutcomeCarboplatin (n=223)Cisplatin (n=242)P valueSMDAKI by ICD code18 (8.1)22 (9.1)0.820.036AKI by 30% decrease in eGFR46 (20.6)107 (44.2)<0.0010.521Neutropenia53 (23.8)63 (26.0)0.640.052Ototoxicity58 (26.0)67 (27.7)0.760.038Emesis25 (11.2)45 (18.6)0.030.208Dehydration21 (9.4)32 (13.2)0.250.120Thrombocytopenia53 (23.8)47 (19.4)0.300.106Peripheral neuropathy69 (30.9)62 (25.6)0.240.118Any symptom based adverse event (rows above this)167 (74.9)208 (86.0)0.0040.282Post-surgery hospitalized ever (surgery +14 through surgery +400)79 (35.4)112 (46.3)0.020.222Mean hospital count0.70±1.291.00±1.690.030.200Minimum eGFR post-treatment start (mL/min/1.73 m2)67.0 [49.0, 81.5]58 [46, 75]0.0010.303Percent change in baseline eGFR (%)−12.4 [−27.0, −1.0]−26.2 [−43.2, −12.5]<0.0010.125Ever visit ED (date after starting chemo up through 270 after)96 (43.0)118 (48.8)0.250.115Chemo failure <0.0010.367 Never switch and complete 4 cycles125 (56.1)103 (38.0) Switched or got 1–3 cycles98 (43.9)150 (62.0)Any adverse outcome 202 (90.6)230 (95.0)0.090.173Mean adverse events2.76±2.053.41±1.950.0010.323Data are presented as n (%), median [interquartile range] or mean ± standard deviation. AKI, acute kidney injury; ED, emergency department; eGFR, estimated glomerular filtration rate; ICD, International Classification of Diseases; SMD, standardized mean difference. Table of safety and tolerability for carboplatin vs. cisplatin adjuvant therapy for unmatched cohort (including switchers) Data are presented as n (%), median [interquartile range] or mean ± standard deviation. AKI, acute kidney injury; ED, emergency department; eGFR, estimated glomerular filtration rate; ICD, International Classification of Diseases; SMD, standardized mean difference. After IPTW adjustment (Table 4), cisplatin remained associated with greater renal toxicity, with higher rates of ≥30% eGFR decline (42.7% vs. 21.3%, P<0.001), lower post-treatment nadir eGFR (44.7 vs. 53.8, P<0.001), and larger percent decline from baseline (−25.4% vs. −12.2%, P<0.001). Overall symptom-based adverse events remained more frequent with cisplatin (85.0% vs. 73.6%, P=0.007), and the mean number of adverse events was higher (3.41 vs. 2.75, P=0.002). Differences in neutropenia, ototoxicity, emesis, dehydration, or thrombocytopenia, and neuropathy were not statistically significant after weighting. There continued to be a higher rate of hospitalization (46.8% vs. 35.2%, P=0.02) and mean hospital count (1.75 vs. 1.25, P=0.02) for patients who received cisplatin as well as a trend towards higher rates of ED visits with the use of cisplatin. Treatment discontinuation remained more common with cisplatin, with only 37.7% completing four cycles without switching compared with 55.5% in the carboplatin group (P<0.001). Table 4Table of safety and tolerability for carboplatin vs. cisplatin adjuvant therapy for IPTW weighted cohort (including switchers)OutcomeCarboplatin (n=465.7)Cisplatin (n=459.4)P valueSMDAKI by ICD code34.2 (7.3)43.8 (9.5)0.420.079AKI by 30% decrease in eGFR99.3 (21.3)196.0 (42.7)<0.0010.470Neutropenia104.6 (22.5)121.4 (26.4)0.350.092Ototoxicity124.7 (26.8)122.6 (26.7)0.980.002Emesis64.1 (13.8)77.6 (16.9)0.430.087Dehydration38.8 (8.3)63.5 (13.8)0.070.175Thrombocytopenia108.3 (23.3)102.5 (22.3)0.820.023Peripheral neuropathy138.8 (29.8)112.2 (24.4)0.220.121Any symptom based adverse event (rows above this)343.0 (73.6)390.5 (85.0)0.0070.283Post-surgery hospitalized ever (surgery +14 through surgery +400)164.0 (35.2)215.0 (46.8)0.020.237Mean hospital count0.67±1.251.02±1.750.020.230Minimum eGFR post-treatment start (mL/min/1.73 m2)71 [53, 86]56 [44, 74]<0.0010.516Percent change in baseline eGFR (%)−12.2 [−27.1, −3.0]−25.4 [−43.1, −12.2]<0.0010.122Ever visit ED (date after starting chemo up through 270 after)196.0 (42.1)227.2 (49.5)0.140.148Chemo failure <0.0010.362 Never switch and complete 4 cycles258.4 (55.5)173.2 (37.7) Switched or got 1–3 cycles207.3 (44.5)286.2 (62.3)Any adverse outcome 417.3 (89.6)434.9 (94.7)0.060.188Mean adverse events2.75±2.093.41±1.950.0020.329Data are presented as n (%), median [interquartile range] or mean ± standard deviation. AKI, acute kidney injury; ED, emergency department; eGFR, estimated glomerular filtration rate; ICD, International Classification of Diseases; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference. Table of safety and tolerability for carboplatin vs. cisplatin adjuvant therapy for IPTW weighted cohort (including switchers) Data are presented as n (%), median [interquartile range] or mean ± standard deviation. AKI, acute kidney injury; ED, emergency department; eGFR, estimated glomerular filtration rate; ICD, International Classification of Diseases; IPTW, inverse probability of treatment weighting; SMD, standardized mean difference. Efficacy outcomes In IPTW-adjusted analyses, cisplatin was associated with numerically improved survival outcomes compared with carboplatin, although differences did not reach statistical significance. At 5 years, overall survival was 61.0% with cisplatin versus 48.8% with carboplatin [HR 1.19 (1.02, 1.40); P=0.17] (Figure 2). In a two-sample survival analysis, assuming 5-year overall survival of 60% versus 51%, 458 patients per group provided 80% power to detect an absolute difference ≥9%, whereas a 5% difference would require 1,474 per group, suggesting that nonsignificant findings are unlikely due to inadequate power for clinically meaningful effects. Similarly, cisplatin was associated with higher recurrence-free survival [63.9% vs. 56.6%, HR 1.32 (1.06, 1.64); P=0.12] (Figure 3). OS by Kaplan-Meier method for carboplatin versus cisplatin adjuvant therapy for IPTW matched cohorts. CI, confidence interval; HR, hazard ratio; IPTW, inverse probability of treatment weighting; OS, overall survival. RFS for carboplatin versus cisplatin adjuvant therapy for IPTW matched cohorts. CI, confidence interval; HR, hazard ratio; IPTW, inverse probability of treatment weighting; RFS, recurrence-free survival. Additional univariate analysis was conducted to compare patients who did not complete their adjuvant cisplatin course, including those who were lost to all chemo follow-up or received subsequent carboplatin (n=150), to patients who completed 4 cycles of carboplatin (n=125). This exploratory analysis was undertaken for hypothesis generation on the potential impact of incomplete cisplatin treatment on survival outcomes for further evaluation of practical clinical consequence of an abbreviated adjuvant chemotherapy course. For patients unable to complete cisplatin therapy, there was a reduction in 5-year OS; from 61.0% to 54.6%. Comparing further between those who did not complete cisplatin and those that did complete the intended carboplatin course, 5-year overall survival continued to favor cisplatin, though with a reduced discrepancy [54.6% vs. 48.8%, HR 1.17 (0.87, 1.58); P=0.49] (Figure 4). This comparison should be interpreted with caution, as it is subject to substantial selection bias and unmeasured confounding associated with patient selection and is therefore hypothesis-generating rather than confirmatory. Univariate OS by Kaplan-Meier method for completion carboplatin course (4 cycles) versus incomplete cisplatin course (<4 cycles and/or switched to carboplatin). CI, confidence interval; HR, hazard ratio; OS, overall survival. Discussion In this retrospective cohort study, we compared the survival and toxicity outcomes, including rates of treatment completion, for cisplatin versus carboplatin-based adjuvant chemotherapy in a large, database of veterans from 2006 to 2020. We used IPTW propensity weighting to compare groups. In both the IPTW matched and unmatched cohorts, cisplatin remained numerically favored for both overall survival (OS) and recurrence-free survival (RFS), though this difference was not statistically significant. We found that cisplatin use was associated with a significantly higher incidence of renal toxicity, overall adverse events, health care utilization and treatment discontinuation compared to carboplatin use. The trend towards improvement in mortality with cisplatin is consistent with previous data (5-8,37), and further lends modest support towards the current practice of using cisplatin as a first line agent. However, the toxicity profile and effects on treatment completion—though intuitive to many clinicians—have been incompletely established through rigorous comparisons between cisplatin and carboplatin. Our findings showing significantly increased renal toxicity, cumulative toxicity burden, symptomatic adverse events, hospitalizations, and treatment discontinuation with the use of cisplatin, all of which have major implications for clinical practice. Specifically, this study highlights the tension between prescribing a therapy with likely mortality benefit at the cost of worsened quality of life, an increasing focus in the field of oncology (38,39). Increased symptomatic adverse events, emergency room visits, and hospitalizations, can represent a significant burden to patients, who may feel that a modest, population-based improvement in mortality does not compensate for increased suffering. This study provides concrete outcomes and statistics to help frame these complex discussions. In particular, in geographically dispersed populations, increased health care utilization (ED visits, hospitalizations) may have heightened effects on quality of life, placing an undue financial and logistical burden on patients. In our veteran cohort, more than 40% of veterans lived greater than 50 miles from the treating facility. Discussing the additional and often hidden “cost” of increased health care engagement is important for informed, individualized patient decision making. Furthermore, treatment-related toxicities may have cascading negative effects. In our study, cisplatin was associated with both significantly increased rates of acute kidney injury as well as severity of renal injury--with higher percent reductions in estimated glomerular filtration rate (eGFR) and lower post treatment eGFR nadirs. Renal toxicity may impair treatment completion and, if persistent, jeopardize eligibility for future therapies (including immunotherapy or targeted therapy), ultimately disrupting longitudinal cancer care. Although chemotherapy remains an important backbone in resected NSCLC management, its historic survival benefit is modest at approximately 4–8.6% (5-7). For this reason, consolidative immunotherapy or targeted therapy are now often added for eligible patients who have successfully completed chemotherapy to deepen survival advantages. Consolidative immunotherapy in PD-L1-positive patients is associated with meaningful improvements in 5-year OS (DFS HR 0.66 (P=0.004) (29), while adjuvant osimertinib and alectinib add substantial survival benefits in both EGFR and ALK-mutated NSCLC- with a 5-year OS 88% vs. 78% (HR 0.49, P<0.001) for osimertinib versus observation (27) and 2 year DFS of 93.8% versus 63.0% (HR of 0.24, P<0.001) for alectinib versus observation (26). Lower chemotherapy completion and more severe renal toxicity may jeopardize patient eligibility for these adjuvant life-prolonging treatments, thus representing a significant risk to overall outcomes. Finally, the endpoint of early treatment discontinuation is a negative factor in oncology as a surrogate for patient experience. And while cisplatin demonstrated numerically superior survival outcomes, exploratory subgroup analyses suggested that the 5-year OS advantage may be diminished in patients who are unable to complete the full intended treatment course. This suggests that, in addition to the burden of morbidity, dose intensity may play a role in cisplatin-associated survival benefit. This consideration may be especially provocative in treatment decisions for elderly or frail patients, who disproportionately experience treatment related toxicities and thus and more likely to “drop-out” due to toxicity. Our study has several notable strengths. Most importantly, our study uses a highly curated VA dataset that has been carefully developed and maintained over multiple years utilizing manual review enhanced by natural language processing. This dataset is unique in its complexity and completeness and has been used in multiple publications to date (34,35,40,41). Secondly, all data is drawn from the VA electronic health record which allows for exhaustive assessments of longitudinal outcomes as it captures care across inpatient, outpatient, pharmacy, laboratory, procedural, and survivorship settings. Similarly, assessment of drug toxicities is accurate and complete because pharmacy, laboratory, inpatient, and outpatient records are integrated in the electronic medical record. Third, our data set includes important patient populations, particularly older patients with comorbid conditions, who are often excluded from clinical trials, as well as patients who live in rural areas, who often do not enroll in trials because of the logistical burden. This makes our results particularly attractive for extrapolation to many real-world oncology practices with similar patient populations. It is worth noting that our study dates (2006 to 2020) predate widespread adoption of adjuvant targeted therapy; however, we believe the study results are relevant in the modern multimodal treatment era which continues to rely on backbone chemotherapy before targeted therapy, thus facing the same chemotherapy selection challenges that we have examined. A final strength of our study is the use of IPTW methodology, which helps mitigate treatment-selection bias between carboplatin and cisplatin while maintaining maximal sample sizes for real-world applicability. By balancing baseline characteristics between our groups, IPTW allowed us to approximate some of the advantages of randomization within a real-world veteran population, enabling comparative effectiveness analysis for cisplatin versus carboplatin, where a true randomized trial would be infeasible. Several limitations of our study should also be acknowledged. First, we did not observe significant differences in ototoxicity or peripheral neuropathy despite these being well-established cisplatin-associated toxicities. This likely reflects limitations in retrospective coding within CDW datasets rather than a true absence of biologic effect. Second, our study lacked an observation-only comparator arm, precluding assessment of the absolute survival benefit of adjuvant chemotherapy generally, or carboplatin specifically, relative to no systemic treatment. Third, this analysis was not designed as a non-inferiority comparison, and therefore lack of statistical significance in survival outcomes (OS and RFS) should not be interpreted as evidence of equivalence. Fourth, as with all retrospective analyses, residual confounding remains possible despite IPTW adjustment, particularly regarding frailty, performance status, clinician treatment selection, and other factors incompletely captured in administrative databases. This limitation is particularly relevant in the exploratory analysis examining cisplatin treatment completion and survival. Additionally, the VHA population is predominantly male and medically complex, potentially limiting generalizability to broader NSCLC populations, including women, younger patients, and those treated in non-integrated health systems. Finally, our cohort predates widespread incorporation of immunotherapy and targeted therapies, which, while allowing clearer attribution of platinum-related toxicity and efficacy, might affect direct extrapolation to contemporary practice as discussed above. Future studies should evaluate platinum selection within modern perioperative and adjuvant multimodality treatment strategies. Conclusions Overall, our study provides clinically actionable findings related to tolerability and treatment delivery through a robust historic data analysis. Cisplatin was consistently associated with greater toxicity burden, increased health care utilization, and lower treatment completion rates. While these findings do not challenge cisplatin’s role as the preferred standard regimen in appropriately selected patients, they support a more individualized approach to platinum selection. For selected older or medically frail patients, carboplatin-based therapy may offer improved tolerability and treatment completion, a trade-off that might outweigh an apparent reduction in efficacy for individualized patients. These results support pragmatic, shared decision-making and underscore the need for prospective studies, especially in the current era of immunotherapy and targeted adjuvant therapy, to better define optimal adjuvant chemotherapy selection. Supplementary The article’s supplementary files as Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0658/prf