Increased Polymicrobial Infection Rates and Associated Mortality in Patients With Chronic Gastrointestinal Disease Undergoing Antibiotic Spacer Placement for Periprosthetic Joint Infection Authors: Julian Wier, Sahil S Telang, Pranit Kumaran, Avinash Iyer, Jay R Lieberman, Nathanael D Heckmann Published: 2026 ABSTRACT AbstractBackgroundEmerging evidence has identified gut dysbiosis, as a contributor to periprosthetic joint infection (PJI) risk. Chronic gastrointestinal (CGI) disease may result in unique causative organisms in PJI. As such, the present study sought to compare the infective organisms of PJI patients presenting with or without CGI disease.MethodsThe Premier Healthcare Database was retrospectively queried to identify patients (≥18 years) who underwent spacer placement for PJI following total hip or knee arthroplasty between January 2016 and December 2023. Multivariable logistic regression models accounting for joint (hip or knee), sex, and Charlson Comorbidity Index were used to calculate the odds of isolating a particular bacterial species in CGI disease. A subanalysis of patients with polymicrobial infections was conducted to determine rates and adjusted odds of inpatient 90-day mortality.ResultsA total of 9756 patients undergoing antibiotic spacer placement for PJI were identified. Of which, 339 (3.5%) had a history of CGI disease and 9417 (96.5%) had no history of CGI disease. When compared to patients without CGI disease, patients with CGI disease were at a 1.6 (95% confidence interval (CI) = 1.1-2.3) and 4.7 (95% CI = 1.3-16.6) times higher adjusted odds of polymicrobial and Clostridial infection. In the cohort of patients with polymicrobial infections, 7.5-times (95% CI = 3.7-15.3) greater adjusted odds of death was observed when compared to patients without CGI disease (14 patients [13.2%] vs 31 patients [1.4%]).ConclusionCGI disease was associated with significantly higher odds of polymicrobial infections and Clostridial species.Level of EvidenceIII.Keywords: Polymicrobial infection, Gastrointestinal disease, Revision arthroplasty, Periprosthetic joint infection Introduction Emerging evidence in the arthroplasty community has identified the gut microbiome as a possible contributor in the development of periprosthetic joint infection (PJI), with dysbiosis of the gut microbiome conferring an elevated risk [[1], [2], [3]]. Current perspectives on PJI propose the existence of a gut–joint axis, whereby intestinal microbiome dysbiosis influences joint health and PJI pathogenesis [1,2]. This theorized gut–joint axis involves impaired intestinal barrier integrity leading to translocation of gut bacteria into systemic circulation [1,2]. Translocation of commensal gut microbes may seed the joint environment, either by indolent infection of native joint tissue or triggering proinflammatory cascades, both potentially contributing to the manifestation of PJI [2,3]. Building on the gut–joint axis, chronic gastrointestinal (CGI) disease processes such as inflammatory bowel disease (IBD) and diverticulosis represent clinically relevant sources of gut dysbiosis, which can compromise mucosal barriers and facilitate bacterial translocation [1,2,[4], [5], [6]]. The inflammatory state caused by CGI disease has been associated with hematogenous seeding presumably due to increased mucosal permeability [7,8]. Severe gut microbiome disruption from prior Clostridium difficile infection has been associated with an independent increase in PJI risk [[9], [10], [11]]. Similarly, patients with IBD, regardless of medication regimen, also exhibit higher rates of PJI [[12], [13], [14]]. Additionally, microbial analysis of PJI isolates with next-generation sequencing has identified the presence of commensal gut bacteria, while serum samples of PJI patients have demonstrated elevated levels of gut permeability biomarkers [2]. Taken together, this evidence reinforces a potential mechanistic link between CGI, gut dysbiosis, and PJI risk. Despite growing evidence supporting the gut–joint axis, the current arthroplasty literature remains limited in directly comparing microbial profiles in PJI patients with and without CGI disease. Existing evidence primarily consists of isolated cases reports [[15], [16], [17]]. As such the present study sought to compare the microbiological profile of PJI patients presenting with comorbid noninfective enterocolitis or diverticulosis to that of PJI patients without underlying CGI disease. We hypothesized that patients with comorbid CGI disease will have a distinct microbial profile with significant enrichment of commensal gut bacteria species such as gram-negative enteric organisms. To provide further clinical context, we also evaluated the interplay between CGI disease, variation in infective organisms, and mortality rates. Material and methods The Premier Healthcare Database (PHD) was retrospectively queried to identify adult patients (aged ≥18 years) who underwent spacer placement for PJI following total hip arthroplasty or total knee arthroplasty (TKA) between January 2016 and December 2023. Cohort selection Patients with a diagnosis of PJI or other deep infection after total joint arthroplasty undergoing antibiotic spacer placement were identified using a combination of using International Classification of Diseases, Tenth Revision (ICD-10) diagnosis/procedural and Current Procedural Terminology codes, as well as direct hospital charge codes (Appendix). Using any preoperative instance of ICD-10–clinical modification codes (ICD-10–clinical modification K50-K52 and K57), patients with comorbid Crohn’s disease, ulcerative colitis, other noninfectious enterocolitis, or diverticulosis (collectively CGI disease) were identified during the index admission and used to divide the cohort into 2 groups. Patients with active infectious gastrointestinal disease such as acute diverticulitis were excluded. These conditions were grouped given low overall rates of pre-existing CGI diseases in the cohort. Identification of infective organisms Using microbial culture and antibiotic susceptibility data available through the PHD, the infective organisms associated with the identified PJIs were reviewed. Culture data obtained during the admission from synovial fluid, tissue, and/or wound fluid specimens were included for analysis. Patients with missing samples or samples labeled as “contaminate” or “possible contaminate” were excluded. Patients with multiple samples with different infective organisms were categorized as polymicrobial infections. Antibiotic susceptibility data were then analyzed to differentiate methicillin-resistant Staphylococcus aureus and vancomycin-resistant Enterococcus from methicillin sensitive S aureus and Enterococcus species, respectively. Using a combination of clinical relevance and the relative frequency of detected organisms, all microbial isolates were categorized into 1 of 15 subgroups (Appendix) [[18], [19], [20]]. Patients with polymicrobial infections were then also classified using this categorization scheme. The rates of infective organisms by CGI subcomponents (Crohn’s disease/ulcerative colitis, other noninfectious enterocolitis, or diverticulosis) were evaluated to ensure that observed findings were not driven by a specific disease (Supplementary Table 1). Study aims The primary goal of this study was to compare the relative rates of microbial subgroups between patients with and without CGI disease. Given the high frequency and observed differences in rates of polymicrobial infections, a subanalysis of patients with polymicrobial infections was conducted to compare rates of 90-day inpatient mortality in patients with and without CGI disease. Statistical analysis Descriptive statistics are reported to describe patient demographics, hospital characteristics, and comorbidities per the Elixhauser system. A Charlson Comorbidity Index (CCI) was also calculated [21]. Between-group differences in categorical and continuous variables were identified via chi-square test and independent t-tests, where appropriate. Multivariable logistic regression models accounting for index procedure type (TKA or total hip arthroplasty), sex, and CCI were used to calculate the relative odds of isolating a particular microbial subgroup if presenting with CGI disease. A similar model was used to compare mortality rates for the subanalysis using Firth's penalized logistic regression. A Benjamin–Hochberg false discovery rate corrected P value <.05 was considered statistically significant for two-sided tests to account for the expected proportion of false discoveries among significant results when comparing the relative frequencies of infective microbial subgroups. For all other analyses, an uncorrected P value of <0.05 was considered significant for two-sided tests. All statistical analyses were performed using Stata (version 17.0; StataCorp LLC, College Station, TX). Results Demographic and patient characteristics In total, 9756 patients undergoing StageOne antibiotic spacer placement for PJI with concurrent culture positive microbial data were identified. Of these, 339 (3.5%) had a history of CGI disease and 9417 (96.5%) had no history of CGI disease. Patients with CGI disease were more likely to be older (69.42 vs 66.76 years, P < .001) and were associated with longer postoperative hospital length of stay (12.94 vs 8.11 days, P < .001). There were no significant differences in the majority of demographic or hospital factors between the 2 cohorts (Table 1). Table 1Demographic data for CGI disease vs non-CGI disease patients.DemographicsChronic gastrointestinal diseaseN = 339No chronic gastrointestinal diseaseN = 9417P valueAverageStandard deviationAverageStandard deviationAge (years)69.4210.8766.7610.96<0.001Length of stay (days)12.9410.828.117.23<0.001NumberPercentNumberPercentTotal knee arthroplasty22064.90%657669.43%0.058Male sex16548.67%503553.16%0.082Race/ethnicity0.235 Asian20.59%971.02% Black195.60%8068.51% Other123.54%3533.73% White29988.20%789083.31% Hispanic72.06%2712.86%Marital status Married18454.28%504053.22%0.138 Other257.37%4825.09% Single13038.35%389541.13%Insurance type Medicare24672.57%619665.42%0.028 Medicaid185.31%6116.45% Private6017.70%226723.94% Other154.42%3433.62%Bed size <100154.42%3753.96%0.031 100-1993610.62%139214.70% 200-2995115.04%185519.59% 300-3994212.39%116112.26% 400-4996218.29%142915.09% ≥50013339.23%320533.84%Urban Yes31091.45%843889.09%0.274Teaching status Yes22466.08%598363.17%0.339Region Midwest8926.25%219923.22%0.561 Northeast7120.94%210422.22% South16949.85%476150.27% West102.95%3533.73%P value <0.05 are indicated in bold. Demographic data for CGI disease vs non-CGI disease patients. P value <0.05 are indicated in bold. Comorbidities Patients with CGI disease had a significantly higher CCI (4.63 vs 3.63, P < .001). Significant differences between the 2 cohorts were observed in 18 of the 34 comorbidities analyzed (Table 2). Table 2Comorbidity data for CGI disease vs non-CGI disease patients.ComorbidityChronic gastrointestinal diseaseN = 339No chronic gastrointestinal diseaseN = 9417P valueAverageStandard deviationAverageStandard deviationCharlson Comorbidity Index4.632.403.632.06<0.001N%N%Congestive heart failure7722.71%145115.32%<0.001Myocardial infarction298.55%5696.01%0.058Arrhythmia10129.79%210722.25%0.001Valvular disease51.47%1601.69%0.753Pulmonary hypertension133.83%2863.02%0.402Chronic pulmonary disease9828.91%213022.49%0.007Peripheral vascular disease5115.04%6126.46%<0.001Hypertension16047.20%497352.51%0.042Complicated hypertension11433.63%231624.45%<0.001Cerebrovascular accident319.14%7588.00%0.467Hemiplegia/paraplegia20.59%410.43%0.673Other neurological disorders3911.50%8408.87%0.103Diabetes, uncomplicated3710.91%122512.93%0.259Diabetes, complicated8525.07%171618.12%0.001Hypothyroidism8625.37%183619.39%0.008Renal failure9728.61%179618.96%<0.001Liver disease3510.32%5245.53%<0.001Chronic peptic ulcer disease82.36%560.59%0.546Blood loss anemia51.47%1821.92%0.546Deficiency anemia5415.93%6306.65%<0.001Coagulopathy4212.39%5886.21%<0.001Venous thromboembolism6318.58%126613.37%0.007Fluid and electrolyte disorders19958.70%292430.87%<0.001Rheumatic disease4212.39%8118.56%0.016Human Immunodeficiency Syndrome/Acquired Immunodeficiency Syndrome (HIV/AIDS)00.00%110.12%0.529Lymphoma72.06%470.50%<0.001Solid tumor102.95%1611.70%0.087Metastatic cancer61.77%580.61%0.010Obesity11533.92%355637.55%0.152Weight loss4112.09%5615.92%<0.001Alcohol abuse102.95%4074.30%0.22Drug abuse195.60%4344.58%0.392Psychoses30.88%1011.07%0.741Depression8825.96%211922.37%0.135P value < 0.05 are indicated in bold. Comorbidity data for CGI disease vs non-CGI disease patients. P value < 0.05 are indicated in bold. Microbial outcomes Rates of PJI due to polymicrobial infection and Clostridium species were more frequently observed in patients with CGI disease, while rates of other Staphylococcus species were lower (Table 3). Of the 248 samples isolated from the polymicrobial infections in patients with CGI disease, infection with gram-negative enteric and Clostridium species were more commonly observed, while lower rates of coagulase negative Staphylococcus were detected however this did not reach significance (Table 4). Similar findings were observed after multivariable adjustment, with CGI conferring higher odds of polymicrobial and Clostridial infection, as well as polymicrobial gram-negative enteric and Clostridial infection; however, lower odds of polymicrobial coagulase negative Staphylococcus infections (Table 5). Table 3Rates of microbial isolates in CGI disease vs non-CGI disease patients.Parent groupOrganismChronic gastrointestinal diseaseN = 339No chronic gastrointestinal diseaseN = 9417P-valueaSkin floraMSSA4613.6%169218.0%0.143Coagulase-negative staphylococci205.9%5355.7%0.927Cutibacterium sp.41.2%1171.2%0.927Corynebacterium sp.51.5%1261.3%0.927Enteric organismsGram-negative enterics sp.308.8%7477.9%0.927Enterococcus sp.92.7%3013.2%0.927Clostridium sp61.8%190.2%0.045Atypical/hospital acquired organismsMRSA319.1%103411.0%0.861Fungal sp.113.2%3393.6%0.927Pseudomonas sp.92.7%2552.7%0.953VRE10.3%160.2%0.587OtherPolymicrobial sp.10631.3%215922.9%0.015Other species278.0%8428.9%0.927Streptococcus sp. other288.3%7498.0%0.927Staphylococcus sp. other61.8%4865.2%0.038MRSA, methicillin-resistant Staphylococcus aureus; MSSA, methicillin-sensitive S aureus; sp., species; VRE, vancomycin-resistant Enterococcus species.P value <0.05 are indicated in bold.aAfter Benjamini–Hochberg adjustment. Rates of microbial isolates in CGI disease vs non-CGI disease patients. MRSA, methicillin-resistant Staphylococcus aureus; MSSA, methicillin-sensitive S aureus; sp., species; VRE, vancomycin-resistant Enterococcus species. P value <0.05 are indicated in bold. After Benjamini–Hochberg adjustment. Table 4Rates of polymicrobial species isolated in CGI disease vs non-CGI disease patients.Parent groupPolymicrobial isolatesChronic gastrointestinal diseaseN = 106 patients with polymicrobial PJINo chronic gastrointestinal diseaseN = 2159 patients with polymicrobial PJIP valueaSkin floraMSSA4138.7%65830.5%0.175Coagulase-negative staphylococci98.5%40618.8%0.070Cutibacterium sp.10.9%361.7%0.467Corynebacterium sp.65.7%1587.3%0.658Enteric organismsGram-negative enterics sp.2927.4%39818.4%0.047Enterococcus sp.1817.0%30714.2%0.569Clostridium sp.98.5%532.5%0.014Atypical/hospital acquired organismsMRSA1110.4%27812.9%0.658Fungal sp.2523.6%41019.0%0.467Pseudomonas sp.65.7%1949.0%0.467VRE43.8%462.1%0.467OtherOther species3835.8%81037.5%0.888Streptococcus sp. other3230.2%51623.9%0.658Staphylococcus sp. other1917.9%38818.0%0.888MRSA, methicillin-resistant Staphylococcus aureus; MSSA, methicillin-sensitive S aureus; sp., species; VRE, vancomycin-resistant enterococcus species.Percentages are reported in events per patient with polymicrobial PJI.P value <0.05 are indicated in bold.aAfter Benjamini–Hochberg adjustment. Rates of polymicrobial species isolated in CGI disease vs non-CGI disease patients. MRSA, methicillin-resistant Staphylococcus aureus; MSSA, methicillin-sensitive S aureus; sp., species; VRE, vancomycin-resistant enterococcus species. Percentages are reported in events per patient with polymicrobial PJI. P value <0.05 are indicated in bold. After Benjamini–Hochberg adjustment. Table 5Adjusted odds of isolating organisms for CGI disease vs non-CGI disease.Infection typeaOR95% CIP valuePolymicrobial1.611.132.29.009Clostridium species4.681.3216.56.017Polymicrobial Clostridium species3.291.566.93.002Polymicrobial gram-negative enteric species1.581.012.47.045Polymicrobial coagulase-negative staphylococci0.430.220.86.017sp, species; aOR, adjusted odds ratio.P value <0.05 are indicated in bold. Adjusted odds of isolating organisms for CGI disease vs non-CGI disease. sp, species; aOR, adjusted odds ratio. P value <0.05 are indicated in bold. The overall 90-day inpatient mortality rate was 2.0% for both the entire cohort and for patients with or without polymicrobial infections. In the cohort of patients with polymicrobial infections, significantly higher rates of mortality were observed in patients with CGI disease (14 patients [13.2%] vs 31 patients [1.4%], P < .001), resulting in 7.5-times (3.7-15.1, P < .001) greater adjusted odds of death than patients without CGI disease. For patients without polymicrobial infections, no significant differences in adjusted odds of death were observed between-study groups (P = .309). Discussion In this study, we investigated the microbial profiles of PJI patients who underwent antibiotic spacer revision arthroplasty, focusing on the influence of gastrointestinal pathology diseases such as IBD and diverticulosis. Our results demonstrate that patients with CGI diseases had significantly higher rates of polymicrobial infections, particularly with gram-negative enteric and Clostridial species, than those without CGI disease. Concurringly, patients with polymicrobial infections and CGI disease were at 7.5-times greater odds of death despite treatment. We hypothesize that the observed shift in microbial composition, with decreased isolation of skin flora-like Staphylococcus epidermidis and increased detection of enteric pathogens, may reflect that these infections are mediated by hematogenous spread of gut-derived microbes rather than direct surgical contamination. Prior data suggest that while distant infectious sources are frequently detected in hematogenous PJIs, underlying predisposing conditions without discrete infectious signs still represent nearly 20% of these cases [22]. Our data provide further evidence for this association, highlighting CGI disease as a potential link between comorbid disease, gut permeability, and species-specific PJI risk profiles. Chisari et al. [2,8] demonstrated elevated zonulin and soluble CD14 levels in patients with PJI, biomarkers associated with increased gut permeability and systemic inflammation. Zonulin levels were particularly elevated in acute PJI, suggesting that loss of gut barrier function may precede or even precipitate infection. Preclinical models have corroborated this gut–joint axis. Methicillin-resistant Staphylococcus aureus–colonized neutrophils harvested from the gut were capable of establishing PJI in naïve animals, providing direct in vivo evidence for the “Trojan horse” mechanism of cell-mediated bacterial dissemination [1,23]. Additionally antibiotic-induced dysbiosis in mice heightens the susceptibility to PJI after local methicillin sensitive S aureus inoculation, highlighting how changes in gut microbiota composition alone can predispose to extra-intestinal infection via hematogenous spread [24]. This gut–joint connection is further reinforced by data demonstrating that gastrointestinal colonization with pathologic bacteria resulted in higher rates of surgical site infections than those exposed via the bloodstream or wound directly [25]. Several studies have demonstrated the association between IBD and PJI risk [4,26,27]. In a database study of over 1.3 million patients undergoing TKA, significantly higher rates of PJI in IBD patients was reported. Notably, the same study also highlighted a 2.6 times greater odds of co-occurring C difficile infection rate in this IBD cohort. This is particularly relevant, given evidence that patients with a history of C difficile infection are at elevated risk of PJI, with those patients undergoing TKA within 3 months of an acute infectious episode at greatest risk [9]. Here, we observed a 4.7-fold increase in the adjusted odds of Clostridial infection, organisms known to proliferate under dysbiotic conditions. Interestingly, we must note that when evaluating the infective organisms across CGI disease subgroups, Clostridial infections were not observed in the IBD group. Others have highlighted how Clostridial and enteric organisms increasingly appear as extraintestinal pathogens in patients with CGI pathology such as IBD, which is commonly associated with concurring pathologies such as barrier breakdown, immunosuppression requirements, and liver disease [[28], [29], [30]]. These findings, combined with the data herein, underscore the importance of heightened vigilance and targeted perioperative strategies in patients with evidence of ongoing gastrointestinal dysbiosis, not just IBD, particularly those with recent Clostridial infections. Polymicrobial PJI represents one of the most recalcitrant and morbid phenotypes in arthroplasty infection. Numerous studies have demonstrated that patients with polymicrobial PJI experience higher rates of surgical failure and mortality than their monomicrobial counterparts [[31], [32], [33]]. Specifically, treatment failure rates have been cited at over 50%, with increased risk of failure in patients with delayed diagnosis [32,33]. While monomicrobial infections typically involve common staphylococcal species, polymicrobial infections can more commonly include a broader spectrum of organisms, such as Enterobacteriaceae, anaerobes, and multidrug-resistant organisms [34,35]. Despite this, several studies on the microbiological profiles of polymicrobial PJIs frequently observed high rates of S epidermidis, which may reflect contaminants or more indolent infection [18,36,37]. Notably, here we observed significantly lower rates of polymicrobial coagulase-negative Staphylococci infections in the CGI disease cohort, potentially reflecting a propensity for more severe pathology and death in this high risk cohort. Conversely, the overrepresentation of gram-negative enterics and Clostridial species may represent enhanced translocation. Disturbance in the gastrointestinal microenvironment in the setting of systemic illness has been associated with increased rates of bacterial translocation, with enterotoxins secreted by gram-negative enteric species further disrupting the intestinal barrier, thus potentially contributing to the probability of polymicrobial hematogenous seeding [38]. These findings also align with others, who noted that coagulase-negative Staphylococci and Enterococci frequently coexist with enteric flora in polymicrobial PJI, and that Clostridium species, though rare, were disproportionately represented in patients with systemic comorbidities [18,39]. There are limitations associated with the present analysis. The use of administrative data is subject to potential misclassification bias. As the PHD only captures in-hospital or clinic visit coded events, the mortality estimates are likely under-reported. Nevertheless, it is unlikely that there are differences in capture rates between the study cohorts. Moreover, although the use of charge-based antibiotic thresholds improved specificity, there remains the possibility that some PJI cases were excluded. Given the low relative rates of CGI disease in this cohort, more robust efforts to control for potential confounding were not feasible. To that end, we must acknowledge the potential for residual confounding which is contributing to the observed effects here and the inability to make causal inferences. In this setting, low event rates such as mortality and Clostridial infections are at risk for instability, thus the data should be interpreted with further caution. Despite the use of multivariable adjustment on demographic, comorbidity, and surgical factors, CGI disease co-occurs with several comorbid conditions and reflects a heterogeneous patient population which likely contributes to the observed findings. Confirming a causal framework for this research question using this retrospective data is not possible, thus these data must be viewed as largely hypothesis generating and to provide awareness of this observed association. Similarly, more specific subanalyses of the unclassified pathogens or the distinct polymicrobial combinations had low power, limiting clinical interpretability. The lack of information related to lab standards for culture processing must also be acknowledged, thus bias related to inaccurate coding of samples, including those labeled as contaminates, must be noted. Treatment failure and reinfection were not directly assessed and are avenues for future study. Finally, varied medical therapy for CGI disease was not accounted for, representing another source of potential confounding. Conclusions This study demonstrates that patients with IBD and diverticulosis undergoing antibiotic spacer placement for PJI exhibit a distinct and concerning microbiological profile. These findings highlight the need for the early identification of infective organisms in this population and support future prospective studies evaluating the impact of expanded empiric antimicrobial coverage to reduce mortality rates. CRediT authorship contribution statement Julian Wier: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Sahil S. Telang: Writing – review & editing, Writing – original draft, Investigation, Formal analysis, Data curation. Pranit Kumaran: Writing – review & editing, Writing – original draft, Investigation. Avinash Iyer: Writing – review & editing, Writing – original draft, Investigation, Data curation. Jay R. Lieberman: Writing – review & editing, Supervision, Conceptualization. Nathanael D. Heckmann: Writing – review & editing, Supervision, Methodology, Conceptualization. Conflicts of interest Jay R. Lieberman receives royalties from DePuy (A Johnson & Johnson company); is a paid consultant for DePuy (A Johnson & Johnson company); holds stock or stock options in BP Surgiphor, Hip innovations Technologies; receives royalties, financial or material support from Saunders/Mosby-Elsevier; and serves as a board or committee member for AAOS, Hip Society, Musculoskeletal Transplant Foundation, Western Orthopaedic Association. Nathanael Heckmann receives royalties from Corin USA; is a paid consultant for Intellijoint Surgical, MicroPort Orthopaedics, Corin USA, Zimmer; holds stock or stock options in Intellijoint Surgical; serves as a board or committee member for AAOS, AJRR, AAHKS, Knee Society, Hip Society. All other authors declare no potential conflicts of interest. For full disclosure statements refer to https://doi.org/10.1016/j.artd.2026.102140. Supplementary data Appendix Conflict of Interest Statement for Lieberman Conflict of Interest Statement for Wier Conflict of Interest Statement for Iyer Conflict of Interest Statement for Kumaran Supplementary Table 1 Conflict of Interest Statement for Heckmann Conflict of Interest Statement for Telang References Associated Data Supplementary Materials Appendix Conflict of Interest Statement for Lieberman Conflict of Interest Statement for Wier Conflict of Interest Statement for Iyer Conflict of Interest Statement for Kumaran Supplementary Table 1 Conflict of Interest Statement for Heckmann Conflict of Interest Statement for Telang