BACKGROUND
Antitumor necrosis factor (TNF) therapy is generally considered first-line treatment of moderate-to-severe Crohn's disease (CD) (1–5 6 7 8,9 10
Additional treatment options include vedolizumab, an anti-α4β7 integrin antibody, and ustekinumab, an anti-interleukin-12/23 antibody. Because anti-TNF refractory patients respond less well to subsequent treatments (11–15 16–20 21,22
METHODS
Study design and data sources
We conducted an active-comparator, new-user (23 24
Study population and exposures of interest
Our cohort included individuals newly initiating ustekinumab or vedolizumab between March 1, 2017, 6 months after the US Food and Drug Administration approved ustekinumab for the treatment of CD, and March 2, 2020. The index date (date of cohort entry) was defined as the date of first qualifying dispensing or administration of ustekinumab or vedolizumab. New use was defined as no evidence of either study drug in the 183 days before the index date (baseline period). All patients were required to be at least 18 years as of the index date and have continuous enrollment in a health plan with medical and drug benefits during the baseline period. Patients were required to have at least 2 diagnoses of CD (International Classification of Diseases, Tenth Revision, Clinical Modification [ICD-10-CM ] diagnosis code K50.XXX) in any care setting and at least 1 dispensing or administration of an anti-TNF inhibitor (adalimumab, certolizumab, golimumab, or infliximab) in the baseline period. Patients were excluded from the cohort if they had 2 or more diagnoses of ulcerative colitis (ICD-10-CM K51.XXX) in any care setting during the baseline period or if they initiated both study drugs on the index date. Patients were allowed to enter the cohort once according to their first qualifying exposure. Ustekinumab, vedolizumab, and anti-TNF inhibitors were identified using National Drug Codes and Healthcare Common Procedure Coding System, level II codes (see Supplementary Table 1, https://links.lww.com/AJG/C751
Outcomes of interest
Effectiveness.
Our primary measure of effectiveness was treatment persistence beyond 52 weeks, evaluated in a cohort that was continuously enrolled in health plans with medical and drug benefits for at least 455 days after cohort entry. Treatment persistence was defined by evidence of a ≥ 1 dispensing or administration of the index drug 365–455 days after the index date. For example, a patient who initiated ustekinumab on March 1, 2017, was said to be persistent to ustekinumab if they had evidence of a refill for ustekinumab between March 1, 2018, and May 30, 2018.
Our secondary measures of effectiveness included (i) all-cause hospitalization, (ii) hospitalization for CD with surgery, and (iii) hospitalization for CD without surgery, all evaluated as time-to-event outcomes over the first year of treatment initiation. Hospitalization for CD with surgery was defined as an inpatient encounter with CD as the principal diagnosis and with evidence of a procedure code for surgery related to CD during that hospitalization. Hospitalization for CD without surgery was an inpatient encounter with CD as the principal diagnosis and without evidence of a procedure code for surgery related to CD during that hospitalization. Surgical procedures were identified using procedure codes specified in a previous publication (25
Safety.
We evaluated 4 safety outcomes, including hospitalizations for (i) any infection, (ii) any malignancy, (iii) a cardiac event, and (iv) a thromboembolic event. Hospitalization for any infection was defined as an inpatient encounter with an ICD-10-CM diagnosis code for a serious or nonserious infection as the principal diagnosis (26 27 28 https://links.lww.com/AJG/C751
Follow-up time
For our primary analysis of treatment persistence, follow-up ended 455 days after cohort entry. Our secondary measures of effectiveness and safety were all evaluated in a cohort that did not enforce any post-index enrollment. For this cohort, follow-up was defined analogous to an intent-to-treat design and began the day after cohort entry and ended at the earliest occurrence of outcome, 365 days after cohort entry, disenrollment, death, or end of data.
Baseline characteristics
Demographic characteristics such as age and sex were ascertained on the index date. Clinical characteristics were evaluated over the baseline period. Measures of general health and CD-related characteristics included (i) Charlson/Elixhauser combined comorbidity index (29 30 Clostridioides difficile testing, (vii) C. difficile diagnosis (31 https://links.lww.com/AJG/C751
Control for confounding
We used propensity score fine stratification (stratum weighting) to control for confounding (32,33 C. difficile testing, inpatient encounter for CD, number of all-cause outpatient encounters, and number of all-cause emergency department encounters. Patients initiating ustekinumab or vedolizumab who had nonoverlapping propensity scores were trimmed (i.e., removed) from the analysis. We created 50 strata based on the distribution of propensity score in patients initiating ustekinumab and then estimated the average treatment effect weights for all patients in strata with variation in treatment. Patients in strata without variation in treatment were dropped from the analysis because they were ineligible for weighting. We used standardized differences to compare the distribution of demographic and clinical characteristics at baseline between new users of ustekinumab and vedolizumab before and after propensity score fine stratification. Differences were deemed to be meaningful if the absolute value of standardized difference was greater than 10% (34
Statistical analysis
For the primary binary outcome of treatment persistence, we estimated risk ratios (RRs) and 95% confidence intervals (CIs) at each data source using a weighted modified Poisson regression model (35 36
Subgroup analyses
We estimated the effects of ustekinumab relative to vedolizumab separately by age (18–34, 35–59, or ≥60 years), recent use of immunosuppressive therapy (yes/no), and number of unique anti-TNF agents used in all available patient history (1 or ≥2). Subgroups were created from the overall trimmed cohorts. Propensity scores and weights were re-estimated within each subgroup before effect estimation.
Sensitivity analysis
Because the cohort used to evaluate our primary measure of treatment persistence required health plan enrollment for 15 months after index treatment initiation, we evaluated the robustness of our findings to potential selection bias by conducting a sensitivity analysis specifying treatment discontinuation as a measure of treatment ineffectiveness. This outcome was evaluated in a cohort that did not enforce any post-index enrollment. Follow-up began the day after cohort entry and ended at the earliest occurrence of treatment discontinuation, 455 days after index treatment initiation, disenrollment, death, or end of data. We truncated the follow-up at a maximum of 455 days to mimic the evaluation window used to ascertain treatment persistence in the primary analysis. Continuous episodes of treatment were constructed from treatment initiation on the index date allowing gaps of 112 days between refills (double the length of a typical regimen). The date of treatment discontinuation was ascertained as the earliest of 56 days after last observed treatment, the date of treatment crossover (ustekinumab users starting vedolizumab or vedolizumab users starting ustekinumab), or the date of dispensing of an anti-TNF inhibitor (adalimumab, certolizumab, golimumab or infliximab). We used propensity score fine stratification to control for confounding and estimated adjusted HRs and 95% CIs as described above. Treatment discontinuation was evaluated overall and by subgroups of age, history of immunosuppressive therapies, and anti-TNF use as described above.
Execution of distributed analysis
Common analytic programs were used at each data partner. Each data partner executed the analytic programs against their data held in a common data model behind institutional firewalls. Data partners shared summary, risk set, and effect estimate data, which were subsequently reviewed and aggregated. All analyses were conducted using the Sentinel Propensity Score Analysis Tool (37
RESULTS
Within the Anthem data, we identified 885 new users of ustekinumab and 490 new users of vedolizumab who met eligibility criteria for our primary analysis. In the Humana data, only 135 ustekinumab and 65 vedolizumab initiators were eligible for primary analysis. Because the sample size was too low to allow for adequate control of confounding or estimate effect size with reasonable precision, the Humana data were considered insufficient for an inferential analysis. Hence, we present results for Anthem alone.
Characteristics of the Anthem study population before and after trimming and propensity score fine stratification are presented in Table 1 . Before propensity score adjustment, the mean age of ustekinumab and vedolizumab initiators at cohort entry was 41.6 (SD 14.6) and 43.8 (SD 14.2) years, respectively. There was a slight female predominance in both treatment groups. Most of the study population had evidence of experience with a single anti-TNF agent in their available claims history before cohort entry. Before propensity score adjustment, 28% ustekinumab initiators vs 25% vedolizumab initiators had experience with 2 or more unique anti-TNF agents in all available history. After weighting, ustekinumab and vedolilzumab initiators seemed balanced on measured baseline characteristics at cohort entry and had overlapping propensity score distributions (Figure 1 ).
Table 1.: Baseline characteristics of adults with Crohn's disease newly initiating treatment with ustekinumab or vedolizumab after failure of treatment with TNF inhibitors before and after propensity score stratum weighting (cohort with 455 days of post-index enrollment)
Figure 1.: Distribution of propensity scores before and after trimming and stratum weighting.
After trimming by propensity score, there were 884 ustekinumab new users, of whom 404 showed persistence beyond 52 weeks (crude risk 45.7 per 100 new users), and 484 new users of vedolizumab, of whom 205 showed persistence beyond 52 weeks (crude risk 42.3 per 100 new users). We observed no difference between ustekinumab and vedolizumab initiators in our primary outcome of treatment persistence beyond 52 weeks (crude RR 1.08 [95% CI 0.91–1.28]; adjusted RR 1.09 [0.95–1.25]; Figure 2 ). Similarly, in our sensitivity analysis of the cohort without a 455-day minimum post-index enrollment requirement, we observed no difference in the outcome of treatment discontinuation (unadjusted HR 0.85 [0.71–1.02]; adjusted HR 0.87 [0.72–1.05]; see Supplementary Table 4, https://links.lww.com/AJG/C751 Figure 2 , there seemed to be no evidence of variation of the treatment effect by any of these factors.
Figure 2.: Relative risk of treatment persistence beyond one year in ustekinumab vs vedolizumab initiators stratified by demographic factors and prior treatment.
Secondary analyses
Within the Anthem data, we identified 1,217 new users of ustekinumab and 667 new users of vedolizumab who met the eligibility criteria for our secondary analyses. Characteristics of the study population for our secondary analyses are given in Table 2 . Overall, this population showed similar characteristics at baseline as the primary cohort. Measured characteristics were well-balanced after propensity score trimming and weighting. We observed lower all-cause hospitalization among ustekinumab initiators vs vedolizumab initiators (adjusted HR 0.73 [0.59–0.91]) and lower nonsurgical CD hospitalizations (adjusted HR 0.58 [0.40–0.83]) (Table 3 ). There was a numerical trend toward lower surgical hospitalizations for CD among ustekinumab new users, but the CIs overlapped the null (adjusted HR 0.83 [0.57–1.22]). Regarding safety outcomes, ustekinumab initiators were less likely to be hospitalized for infection (adjusted HR 0.56 [0.34–0.92]). We observed no statistically significant differences in hospitalization for thrombotic events, although the absolute number of thrombotic events was low with fewer than 10 events identified across groups. The low number of hospitalizations for cardiac events and malignancy precluded meaningful comparisons.
Table 2.: Baseline characteristics of adults with Crohn's disease newly initiating treatment with ustekinumab or vedolizumab after failure of treatment with TNF inhibitors before and after propensity score stratum weighting (cohort without post-index enrollment requirement)
Table 3.: Incidence and effect estimates for secondary end points in new users of ustekinumab vs vedolizumab (cohort without post-index enrollment requirement)
DISCUSSION
In this rigorous pharmacoepidemiologic study, we compared the effectiveness and safety of ustekinumab and vedolizumab among anti-TNF-experienced patients with CD in a large, geographically diverse real-world population. We observed no differences in our primary outcome of treatment persistence beyond 52 weeks; however, initiators of ustekinumab experienced fewer total hospitalizations, nonsurgical CD hospitalizations, and hospitalizations for infection.
Several recent European studies have sought to compare vedolizumab with ustekinumab in this patient population (16–22 18 21 22
Strengths of this study include the large and geographically diverse study population allowing for the generation of real-world evidence and the use of a rigorous pharmacoepidemiologic study design to minimize bias and control for confounding to the greatest extent possible. As with any study that relies on health insurance claims data, there is always the possibility of misclassification. By requiring several diagnosis codes for CD along with treatment codes for both anti-TNF inhibitors and either ustekinumab or vedolizumab and by excluding patients with concomitant diagnosis codes for ulcerative colitis, we do not expect significant misclassification of CD itself. Furthermore, because we ascertained treatment exposure and persistence using specific codes for medication dispensing and/or administration, misclassification of treatment status is also unlikely. Although the outcome of all-cause hospitalization is also likely to be quite valid, we acknowledge that cause-specific hospitalization (e.g. CD, infection) may be less accurate and that any misclassification is likely to be at random and result in bias toward the null. An additional limitation is that claims data lack clinical details such as CD phenotype and severity and lifestyle factors such as smoking status. Thus, although we were able to adjust for a number of factors related to comorbidities, health services utilization, and surrogate markers of disease activity such as anemia and malnutrition, we recognize the possibility of unmeasured confounding. To help quantify the potential for unmeasured confounding, we compared the characteristics of ustekinumab and vedolizumab users in a separate prospective multicenter US cohort, Study of a Prospective Adult Research Cohort with inflammatory bowel disease, and found initiators of both treatments to be relatively comparable across a broad range of demographic and clinical characteristics (see Supplementary Table 5, https://links.lww.com/AJG/C751 a priori with input from our patient coinvestigators because this was considered a balancing measure reflecting both effectiveness and safety. However, we recognize that treatment persistence is an indirect end point, and future research looking at clinical outcomes including disease complications and specific safety events is needed. Indeed, our findings of differences in between treatment groups in secondary outcomes, but not our primary outcome, suggests that future comparative effectiveness research studies may wish to specify primary end points other than persistence. We also did not collect data on steroid tapering/refills, and this can be another outcome in future studies. We acknowledge that the anti-TNF-experienced patients in our study are a heterogeneous population, including those who used anti-TNF inhibitors for variable time periods and those who stopped their TNF inhibitors for reasons related to safety, lack of effectiveness, loss of effectiveness, and other reasons. Owing to the lack of granularity in claims data and the fact that many Americans change health insurance every few years, analyses of how these factors affect the effectiveness of subsequent biologic therapy are beyond the scope of this report. In addition, our study included commercially insured adults continuously enrolled in their health plans in the United States, limiting generalizability to low-income patients (without insurance or insured by Medicaid), older patients (with Medicare insurance), and children.
In conclusion, this real-world comparative effectiveness study of anti-TNF-experienced patients with CD initiating vedolizumab or ustekinumab showed similar treatment persistence rates beyond 52 weeks, although many secondary outcomes such as all-cause hospitalization, nonsurgical CD hospitalization, and hospitalization for infection favored ustekinumab initiation. We, therefore, advocate for individualized decision making in this medically refractory population, considering patient preference and other factors such as cost and route of administration.
CONFLICTS OF INTEREST
Guarantor of the article: Michael D. Kappelman, MD, MPH.
Specific author contributions: M.D.K.: conception, funding acquisition, investigation, methodology, and writing of the original draft. J.B.: funding acquisition; investigation; and writing, review, and editing. J.E.D.: funding acquisition; investigation; and writing, review, and editing. A.D.: funding acquisition; investigation; and writing, review, and editing. J.D.L.: conception; funding acquisition; investigation; and writing, review, and editing. M.D.L.: conception; funding acquisition; investigation; and writing, review, and editing. All other authors: investigation and writing, review, and editing.
Financial support: The research reported in this article was funded through a Patient-Centered Outcomes Research Institute (PCORI) Award (PaCR-2017C2-8172-IC). The statements and opinions in this article are solely the responsibility of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute (PCORI) or its Board of Governors or Methodology Committee.
Potential competing interests: M.D.K. has consulted for AbbVie, Janssen, Pfizer, Takeda, and Lilly; is a shareholder in Johnson & Johnson; and has received research support from Pfizer, Takeda, Janssen, AbbVie, Lilly, Genentech, Boehringer-Ingelheim, Bristol-Myers Squibb, Celtrion, and Arenapharm. S.A., L.H., A.E.W., S.S., A.L.S., J.S.R., S.T. are all employees of the Harvard Pilgrim Health Care Institute. The Harvard Pilgrim Health Care Institute is a nonprofit organization that conducts work for government and private organizations, including pharmaceutical companies. S.T. consults for Pfizer Inc. and Merck Co. on methodological projects unrelated to the submitted work. L.P. is an employee of Anthem, Inc. She has received research support from Sanofi on a project unrelated to the submitted work. E.M. is an employee of StatLog Inc.; StatLog has received consultancy fees from the Harvard Pilgrim Health Care Institute. V.N. was an employee of and had stock/stock options in Humana Healthcare Research, a wholly owned subsidiary of Humana, Inc., at the time of this work. J.E.D. is a shareholder in Pfizer. K.H. was an employee of and had stock/stock options in Anthem, Inc. at the time of this work. He is currently an employee of Janssen Research & Development. A.D. reports no conflicts of interest. J.D.L. consulted or served on an advisory board for Eli Lilly and company, Samsung Bioepis, UCB, Bristol-Myers Squibb, Nestle Health Science, Merck, Celgene, Janssen Pharmaceuticals, Bridge Biotherapeutics, Entasis Therapeutics, AbbVie, Pfizer, Gilead, Arena Pharmaceuticals, Protagonist Therapeutics, Amgen, and Scipher Medicine. He has had research funding from Nestle Health Science, Takeda, Janssen Pharmaceuticals, and AbbVie. He has performed legal work on behalf of generic manufacturers of ranitidine, including L. Perrigo Company; Glenmark Pharmaceuticals Inc.; Amneal Pharmaceuticals LLC; Aurobindo Pharma USA, Inc.; Dr. Reddy's Laboratories, Inc.; Novitium Pharma; Ranbaxy Inc.; Sun Pharmaceutical Industries, Inc.; Strides Pharma, Inc.; and Wockhardt USA LLC. He owns stock in Dark Canyon Labs. M.D.L. has consulted for AbbVie, Janssen, Pfizer, Takeda, Lilly, BMS, Genentech, Roche, Target Pharmasolutions, Prometheus, Calibr, and Theravance and has received research support from Pfizer and Takeda.Study Highlights
WHAT IS KNOWN
✓ Many patients with Crohn's disease (CD) lose response or become intolerant to anti-TNF therapy.
✓ Subsequent treatment options include ustekinumab and vedolizumab.
✓ There is a paucity of research comparing the effectiveness of these treatments in anti-TNF-experienced patients with CD, and existing studies have yielded contradictory results.
WHAT IS NEW HERE
✓ We compared the effectiveness and safety of ustekinumab to vedolizumab in a large, geographically diverse US population of TNF-experienced patients with CD and found no difference in our primary outcome of treatment persistence beyond 1 year.
✓ Secondary outcomes of all-cause hospitalization, nonsurgical CD hospitalization, and hospitalization for infection favored ustekinumab initiation.
✓ We advocate for individualized decision making in this medically refractory population.
REFERENCES
1. Gomollon F, Dignass A, Annese V, et al. 3rd European evidence-based consensus on the diagnosis and management of Crohn's disease 2016: Part 1: Diagnosis and medical management. J Crohns Colitis 2017;11(1):3–25.
2. Dassopoulos T, Sultan S, Falck-Ytter YT, et al. American Gastroenterological Association Institute technical review on the use of thiopurines, methotrexate, and anti-TNF-alpha biologic drugs for the induction and maintenance of remission in inflammatory Crohn's disease. Gastroenterology 2013;145(6):1464–78.e5.
3. Harbord M, Eliakim R, Bettenworth D, et al. Third European evidence-based consensus on diagnosis and management of ulcerative colitis. Part 2: Current management. J Crohns Colitis 2017;11(7):769–84.
4. Dassopoulos T, Cohen RD, Scherl EJ, et al. Ulcerative colitis care pathway. Gastroenterology 2015;149(1):238–45.
5. Bressler B, Marshall JK, Bernstein CN, et al. Clinical practice guidelines for the medical management of nonhospitalized ulcerative colitis: The toronto consensus. Gastroenterology 2015;148(5):1035–58 e3.
6. Burisch J, Kiudelis G, Kupcinskas L, et al. Natural disease course of Crohn's disease during the first 5 years after diagnosis in a European population-based inception cohort: An epi-IBD study. Gut 2019;68(3):423–33.
7. Feuerstein JD, Ho EY, Shmidt E, et al. AGA clinical practice guidelines on the medical management of moderate to severe luminal and perianal fistulizing Crohn's disease. Gastroenterology 2021;160(7):2496–508.
8. Billioud V, Sandborn WJ, Peyrin-Biroulet L. Loss of response and need for adalimumab dose intensification in Crohn's disease: A systematic review. Am J Gastroenterol 2011;106(4):674–84.
9. Karmiris K, Paintaud G, Noman M, et al. Influence of trough serum levels and immunogenicity on long-term outcome of adalimumab therapy in Crohn's disease. Gastroenterology 2009;137(5):1628–40.
10. Schnitzler F, Fidder H, Ferrante M, et al. Long-term outcome of treatment with infliximab in 614 patients with Crohn's disease: Results from a single-centre cohort. Gut 2009;58(4):492–500.
11. Feagan BG, Rubin DT, Danese S, et al. Efficacy of vedolizumab induction and maintenance therapy in patients with ulcerative colitis, regardless of prior exposure to tumor necrosis factor antagonists. Clin Gastroenterol Hepatol 2017;15(2):229–39.e5.
12. Sands BE, Sandborn WJ, Van Assche G, et al. Vedolizumab as induction and maintenance therapy for Crohn's disease in patients naive to or who have failed tumor necrosis factor antagonist therapy. Inflamm Bowel Dis 2017;23(1):97–106.
13. Feagan BG, Sandborn WJ, Gasink C, et al. Ustekinumab as induction and maintenance therapy for Crohn's disease. New Engl J Med 2016;375(20):1946–60.
14. Sandborn WJ, van Assche G, Reinisch W, et al. Adalimumab induces and maintains clinical remission in patients with moderate-to-severe ulcerative colitis. Gastroenterology 2012;142(2):257–65.e3.
15. Sandborn WJ, Su C, Sands BE, et al. Tofacitinib as induction and maintenance therapy for ulcerative colitis. New Engl J Med 2017;376(18):1723–36.
16. Townsend T, Razanskaite V, Dodd S, et al. Comparative effectiveness of ustekinumab or vedolizumab after one year in 130 patients with anti-TNF-refractory Crohn's disease. Aliment Pharmacol Ther 2020;52(8):1341–52.
17. Alric H, Amiot A, Kirchgesner J, et al. The effectiveness of either ustekinumab or vedolizumab in 239 patients with Crohn's disease refractory to anti-tumour necrosis factor. Aliment Pharmacol Ther 2020;51(10):948–57.
18. Parrot L, Dong C, Carbonnel F, et al. Systematic review with meta-analysis: The effectiveness of either ustekinumab or vedolizumab in patients with Crohn's disease refractory to anti-tumour necrosis factor. Aliment Pharmacol Ther 2022;55(4):380–8.
19. Biemans VBC, van der Woude CJ, Dijkstra G, et al. Ustekinumab is associated with superior effectiveness outcomes compared to vedolizumab in Crohn's disease patients with prior failure to anti-TNF treatment. Aliment Pharmacol Ther 2020;52(1):123–34.
20. Manlay L, Boschetti G, Pereira B, et al. Comparison of short- and long-term effectiveness between ustekinumab and vedolizumab in patients with Crohn's disease refractory to anti-tumour necrosis factor therapy. Aliment Pharmacol Ther 2021;53(12):1289–99.
21. Lenti MV, Dolby V, Clark T, et al. A propensity score-matched, real-world comparison of ustekinumab vs vedolizumab as a second-line treatment for Crohn's disease. The Cross Pennine study II. Aliment Pharmacol Ther 2022;55(7):856–66.
22. Onali S, Pugliese D, Caprioli FA, et al. An objective comparison of vedolizumab and ustekinumab effectiveness in Crohn's disease patients' failure to TNF-alpha inhibitors. Am J Gastroenterol 2022;117(8):1279–87.
23. Lund JL, Richardson DB, Stürmer T. The active comparator, new user study design in pharmacoepidemiology: Historical foundations and contemporary application. Curr Epidemiol Rep 2015;2(4):221–8.
25. Jones DW, Finlayson SRG. Trends in surgery for Crohn's disease in the era of infliximab. Ann Surg 2010;252(2):307–12.
26. Lo Re V III, Carbonari DM, Jacob J, et al. Validity of ICD-10-CM diagnoses to identify hospitalizations for serious infections among patients treated with biologic therapies. Pharmacoepidemiol Drug Saf 2021;30(7):899–909.
27. Poizeau F, Nowak E, Kerbrat S, et al. Association between early severe cardiovascular events and the initiation of treatment with the anti-interleukin 12/23p40 antibody ustekinumab. JAMA Dermatol 2020;156(11):1208–15.
28. Tamariz L, Harkins T, Nair V. A systematic review of validated methods for identifying venous thromboembolism using administrative and claims data. Pharmacoepidemiol Drug Saf 2012;21(Suppl 1):154–62.
29. Sun JW, Rogers JR, Her Q, et al. Adaptation and validation of the combined comorbidity score for ICD-10-CM. Med Care 2017;55(12):1046–51.
30. Cheng AG, Oxford EC, Sauk J, et al. Impact of mode of delivery on outcomes in patients with perianal Crohn's disease. Inflamm Bowel Dis 2014;20:1391–8.
31. Ma GK, Brensinger CM, Wu Q, et al. Increasing incidence of multiply recurrent Clostridium difficile infection in the United States: A cohort study. Ann Intern Med 2017;167(3):152–8.
32. Desai RJ, Rothman KJ, Bateman BT, et al. A propensity-score-based fine stratification approach for confounding adjustment when exposure is infrequent. Epidemiology 2017;28(2):249–57.
33. Desai RJ, Franklin JM. Alternative approaches for confounding adjustment in observational studies using weighting based on the propensity score: A primer for practitioners. BMJ 2019;367:l5657.
34. Austin PC. Using the standardized difference to compare the prevalence of a binary variable between two groups in observational research. Commun Stat Simulation Comput 2009;38(6):1228–34.
36. Shu D, Yoshida K, Fireman BH, et al. Inverse probability weighted Cox model in multi-site studies without sharing individual-level data. Stat Methods Med Res 2020;29(6):1668–81.
