Welcome to visit Zhongnan Medical Journal Press Series journal website!
Zhongnan Medical Journal Press Series of Journals Center for Evidence-Based and Translational Medicine of Wuhan University

The title
  • The title
  • The keywords
  • The author
Submit
Home Articles Vol 35,2026 No.10 Detail

Mechanisms, identification and management of sparse-data bias in pharmacoepidemiologic studies

Published on Nov. 07, 2026Total Views: 34 times Total Downloads: 12 times Download Mobile

Author: GAO Xin 1, 2, 3 GONG Shuo 3 DENG Huiwen 1 HUANG Yue 4 ZHAN Siyan 1, 2, 3, 5 PEI Zhengcun 4 LIU Zhike 1, 2, 3

Affiliation: 1.Department of Epidemiology and Biostatistics, School of Public Health, Peking University, Beijing 100191, China 2.Key Laboratory of Epidemiology of Major Diseases(Peking University), Beijing 100191, China 3.Peking university-Yinzhou Regional joint institute for Advanced Medical Intelligence and Digital health, Ningbo 315100, Zhejiang Province, China 4.Medical College, Tianjin University, Tianjin 300072, China 5.Research Center of Clinical Epidemiology, Peking University Third Hospital, Beijing 100191, China

Keywords: Pharmacoepidemiology Sparse-data bias Separation Penalized likelihood Bayesian methods

DOI: 10.12173/j.issn.1005-0698.202609095

Reference: GAO Xin, GONG Shuo, DENG Huiwen, et al. Mechanisms, identification and management of sparse-data bias in pharmacoepidemiologic studies[J]. Chinese Journal of Pharmacoepidemiology, 2026, 35(10): 1206-1214.[Article in Chinese]

  • Abstract
  • Full-text
  • References
Abstract

Pharmacoepidemiologic studies often use large-scale claims data, electronic health records and registries. However, rare exposures or outcomes, special populations, short risk windows and high-dimensional adjustment may leave insufficient information for estimating a specific effect. Sparse-data bias can cause maximum-likelihood estimates to deviate from the true effect, affecting the interpretation of safety signals and treatment effects. This review describes the concepts and mechanisms of sparse-data bias in common pharmacoepidemiologic settings, distinguishes it from separation, positivity violations and imprecision, and evaluates sparsity and its implications through data structure, estimation and inference diagnostics, and alternative models and sensitivity analyses. It compares the assumptions and limitations of design optimization, Firth bias reduction, data augmentation, propensity-score approaches, and Bayesian methods. Management should begin by defining the estimand and assessing its data support, then improve the design and covariate structure, and select an estimation method appropriate to the source of sparsity. Reporting key counts, penalty or prior specifications and sensitivity analyses helps readers assess the extent to which conclusions depend on the observed data and modeling assumptions.

Full-text
Please download the PDF version to read the full text: download
References

1. ActonEK, WillisAW, HennessyS. Core concepts in pharma-coepidemiology: key biases arising in pharmacoepidemiologic studies[J]. Pharmacoepidemiol Drug Saf, 2023, 32(1): 9-18. DOI: 10.1002/pds.5547.

2. GreenlandS, SchwartzbaumJA, FinkleWD. Problems due to small samples and sparse data in conditional logistic regression analysis[J]. Am J Epidemiol, 2000, 151(5): 531-539. DOI: 10.1093/oxfordjournals.aje.a010240.

3. GreenlandS, MansourniaMA, AltmanDG. Sparse data bias: a problem hiding in plain sight[J]. BMJ, 2016, 352: i1981. DOI: 10.1136/bmj.i1981.

4. 中国药学会. 药物流行病学研究方法学指南(第2版): T/CPHARMA 002-2025[S]. 北京: 中国药学会, 2025.

5. MansourniaMA, GeroldingerA, GreenlandS, et al. Separation in logistic regression: causes, consequences, and control[J]. Am J Epidemiol, 2018, 187(4): 864-870. DOI: 10.1093/aje/kwx299.

6. ZhuY, HubbardRA, ChubakJ, et al. Core concepts in pharmacoepidemiology: violations of the positivity assumption in the causal analysis of observational data: consequences and statistical approaches[J]. Pharmacoepidemiol Drug Saf, 2021, 30(11): 1471-1485. DOI: 10.1002/pds.5338.

7. PetersenML, PorterKE, GruberS, et al. Diagnosing and responding to violations in the positivity assumption[J]. Stat Methods Med Res, 2012, 21(1): 31-54. DOI: 10.1177/0962280210386207.

8. WyssR, van der LaanM, GruberS, et al. Targeted learning with an undersmoothed LASSO propensity score model for large-scale covariate adjustment in health-care database studies[J]. Am J Epidemiol, 2024, 193(11): 1632-1640. DOI: 10.1093/aje/kwae023.

9. GoshoM, OhigashiT, NagashimaK, et al. Bias in odds ratios from logistic regression methods with sparse data sets[J]. J Epidemiol, 2023, 33(6): 265-275. DOI: 10.2188/jea.JE20210089.

10. LanganSM, SchmidtSAJ, WingK, et al. The reporting of studies conducted using observational routinely collected health data statement for pharmacoepidemiology (RECORD-PE)[J]. BMJ, 2018, 363: k3532. DOI: 10.1136/bmj.k3532.

11. PeduzziP, ConcatoJ, KemperE, et al. A simulation study of the number of events per variable in logistic regression analysis[J]. J Clin Epidemiol, 1996, 49(12): 1373-1379. DOI: 10.1016/S0895-4356(96)00236-3.

12. van SmedenM, MoonsKGM, de GrootJAH, et al. Sample size for binary logistic prediction models: beyond events per variable criteria[J]. Stat Methods Med Res, 2019, 28(8): 2455-2474. DOI: 10.1177/0962280218784726.

13. McCaffreyDF, GriffinBA, AlmirallD, et al. A tutorial on propensity score estimation for multiple treatments using generalized boosted models[J]. Stat Med, 2013, 32(19): 3388-3414. DOI: 10.1002/sim.5753.

14. HamraGB, MacLehoseRF, ColeSR. Sensitivity analyses for sparse-data problems—using weakly informative Bayesian priors[J]. Epidemiology, 2013, 24(2): 233-239. DOI: 10.1097/EDE.0b013e318280db1d.

15. GoshoM, IshiiR, NagashimaK, et al. Determining the prior mean in Bayesian logistic regression with sparse data: a nonarbitrary approach[J]. J R Stat Soc Ser C Appl Stat, 2025, 74(1): 126-141. DOI: 10.1093/jrsssc/qlae048.

16. ChoiL, CarrollRJ, BeckC, et al. Evaluating statistical approaches to leverage large clinical datasets for uncovering therapeutic and adverse medication effects[J]. Bioinformatics, 2018, 34(17): 2988-2996. DOI: 10.1093/bioinformatics/bty306.

17. FaksováK, LaksafossAD, HviidA. Human papillomavirus nonavalent (HPV9) vaccination and risk of immune mediated diseases, myocarditis, pericarditis, and thromboembolic outcomes in Denmark: self-controlled case series study[J]. BMJ Med, 2024, 3(1): e000854. DOI: 10.1136/bmjmed-2024-000854.

18. SchultzeA, BrownJ, LogieJ, et al. Overcoming time-varying confounding in self-controlled case series with active comparators: application and recommendations[J]. Am J Epidemiol, 2025, 194(1): 220-225. DOI: 10.1093/aje/kwae216.

19. LiF, MorganKL, ZaslavskyAM. Balancing covariates via propensity score weighting[J]. J Am Stat Assoc, 2018, 113(521): 390-400. DOI: 10.1080/01621459.2016.1260466.

20. TennantPWG, MurrayEJ, ArnoldKF, et al. Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations[J]. Int J Epidemiol, 2021, 50(2): 620-632. DOI: 10.1093/ije/dyaa213.

21. TalbotD, MassambaVK. A descriptive review of variable selection methods in four epidemiologic journals: there is still room for improvement[J]. Eur J Epidemiol, 2019, 34(8): 725-730. DOI: 10.1007/s10654-019-00529-y.

22. RassenJA, GlynnRJ, BrookhartMA, et al. Covariate selection in high-dimensional propensity score analyses of treatment effects in small samples[J]. Am J Epidemiol, 2011, 173(12): 1404-1413. DOI: 10.1093/aje/kwr001.

23. MehtaCR, PatelNR. Exact logistic regression: theory and examples[J]. Stat Med, 1995, 14(19): 2143-2160. DOI: 10.1002/sim.4780141908.

24. HeinzeG, SchemperM. A solution to the problem of separation in logistic regression[J]. Stat Med, 2002, 21(16): 2409-2419. DOI: 10.1002/sim.1047.

25. FirthD. Bias reduction of maximum likelihood estimates[J]. Biometrika, 1993, 80(1): 27-38. DOI: 10.1093/biomet/80.1.27.

26. PuhrR, HeinzeG, NoldM, et al. Firth's logistic regression with rare events: accurate effect estimates and predictions?[J]. Stat Med, 2017, 36(14): 2302-2317. DOI: 10.1002/sim.7273.

27. GreenlandS, MansourniaMA. Penalization, bias reduction, and default priors in logistic and related categorical and survival regressions[J]. Stat Med, 2015, 34(23): 3133-3143. DOI: 10.1002/sim.6537.

28. DiscacciatiA, OrsiniN, GreenlandS. Approximate Bayesian logistic regression via penalized likelihood by data augmentation[J]. Stata J, 2015, 15(3): 712-736. DOI: 10.1177/1536867X1501500306.

29. AustinPC, StuartEA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies[J]. Stat Med, 2015, 34(28): 3661-3679. DOI: 10.1002/sim.6607.

30. GhoshJ, LiY, MitraR. On the use of Cauchy prior distributions for Bayesian logistic regression[J]. Bayesian Anal, 2018, 13(2): 359-383. DOI: 10.1214/17-BA1051.

31. GelmanA, JakulinA, PittauMG, et al. A weakly informative default prior distribution for logistic and other regression models[J]. Ann Appl Stat, 2008, 2(4): 1360-1383. DOI: 10.1214/08-AOAS191.

32. MansourniaMA, NazemipourM. Recommendations for accurate reporting in medical research statistics[J]. Lancet, 2024, 403(10427): 611-612. DOI: 10.1016/S0140-6736(24)00139-9.

33. U.S. Food and Drug Administration. M14 general principles on planning, designing, analyzing, and reporting of non-interventional studies that utilize real-world data for safety assessment of medicines: guidance for industry[EB/OL]. (2026-03)[2026-09-15]. https://www.fda.gov/media/179795/download.

Popular papers
Last 6 months