Comparison of two-stage methods for count data in Mendelian randomization: a simulation study
Publication: Journal of Comparative Effectiveness Research
Abstract
Aim: Mendelian randomization (MR) is an instrumental variable (IV) method that utilizes genetic variants to establish causality between risk factors and outcomes in observational studies. These methods were primarily developed under assumptions appropriate for continuous or approximately normally distributed variables. However, in many biomedical and clinical studies, exposures and outcomes are naturally recorded as counts, such as the number of disease episodes or clinical events. Despite this, two-stage MR methods are applied to count data without a clear understanding of their validity under such settings. While individual-level MR methods like two-stage predictor substitution (TSPS) and two-stage residual inclusion (TSRI) are common, their comparative performance for count exposures and outcomes remains unclear. Materials & methods: We conducted the first systematic evaluation of TSPS and TSRI for count data using Poisson and negative binomial models across realistic MR scenarios. Simulations varied instrument strength (IS), confounding, sample size and also focused on invalid instruments. Performance was assessed by bias, root mean square error (RMSE), 95% confidence interval (CI) coverage, CI width and Type I error rate. To demonstrate practical application, we applied these methods to investigate the causal relationship between alcohol consumption and gout attacks using empirical data. Results: Our results revealed that across all scenarios, TSRI with the Poisson model produced the most stable estimates with lower bias and RMSE. TSRI achieved near-nominal coverage and narrower CI across varying IS and confounding levels, maintaining Type I error close to 0.05. IS significantly impacted performance, with IS = 0.5 yielding estimates closer to the true values, while weaker instruments (IS = 0.1) led to higher RMSE and bias. Increasing sample size in the presence of invalid and weak genetic variants increased the bias. In additional simulations with multiple weak instruments, TSRI continued to outperform TSPS, yielding lower bias or RMSE, narrow CI width and near-nominal coverage across sample sizes. In our application, alcohol consumption was causally associated with an estimated 11.6–12.7% increase in the expected number of gout attacks per year per unit increase in alcohol intake, although the presence of an invalid single nucleotide polymorphism likely biased this estimate. Conclusion: This study advances MR methodology by clarifying how TSPS and TSRI behave with count exposures and outcomes, providing practical guidance for valid instrument selection and reliable causal inference in MR studies involving count data.
Plain language summary
What is this article about?
This study compares two methods used to study whether an exposure causes an outcome, especially when the outcome is a count (like the number of gout attacks). These methods use genetics to help understand cause-and-effect relationships in health research. The goal was to see which method works better when dealing with this kind of data.
What were the results?
Using simulations, we found that the two stage residual inclusion method with a Poisson model gave more accurate results, especially when strong genetic variants were used. When the genetic variants were weak or not valid, the results became more biased. In a real-world example, we looked at whether alcohol consumption causes gout attacks. We found a link but also showed that using an invalid genetic variant could lead to misleading conclusions.
What do the results mean?
This study shows that choosing strong and valid genetic variants is very crucial when using these methods to study cause and effects. Using weak or invalid ones can produce biased results, even if a large amount of data is available. This is important for researchers using genetic data to make health-related decisions.
Supplementary Material
File (supplementary materials.docx)
- Download
- 39.86 KB
References
Papers of special note have been highlighted as: • of interest
1.
Zabor EC, Kaizer AM, Hobbs BP. Randomized controlled trials. Chest 158 (Suppl.), S79–S87 (2020).
2.
Klungel OH, Martens EP, Psaty BM et al. Methods to assess intended effects of drug treatment in observational studies are reviewed. J. Clin. Epidemiol. 57(12), 1223–1231 (2004).
3.
Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G. Mendelian randomization: using genes as instruments for making causal inferences in epidemiology. Stat. Med. 27(7), 1133–1163 (2008).
4.
Evans DM, Davey Smith G. Mendelian randomization: new applications in the coming age of hypothesis-free causality. Annu. Rev. Genomics Hum. Genet. 16, 327–350 (2015).
5.
Baiocchi M, Cheng J, Small DS. Instrumental variable methods for causal inference. Stat. Med. 33(13), 2297–2340 (2014).
6.
Hariton E, Locascio JJ. Randomised controlled trials - the gold standard for effectiveness research: study design: randomised controlled trials. BJOG 125(13), 1716–1716 (2018).
7.
Casanova F, O'Loughlin J, Karageorgiou V et al. Effects of physical activity and sedentary time on depression, anxiety and well-being: a bidirectional Mendelian randomisation study. BMC Med. 21, 123 (2023).
8.
Smith MC, O'Loughlin J, Karageorgiou V et al. The genetics of falling susceptibility and identification of causal risk factors. Sci Rep. 13, 4521 (2023).
9.
Walker V, Sanderson E, Levin MG, Damraurer SM, Feeney T, Davies NM. Reading and conducting instrumental variable studies: guide, glossary, and checklist. BMJ 387, e078093 (2024).
10.
Terza JV, Basu A, Rathouz PJ. Two-stage residual inclusion estimation: addressing endogeneity in health econometric modeling. J. Health Econ. 27(3), 531–543 (2008).
• Introduces the two-stage residual inclusion method in the context of health data and explains its advantages over two-stage predictor substitution in nonlinear settings.
11.
Wilms R, Mäthner E, Winnen L, Lanwehr R. Omitted variable bias: a threat to estimating causal relationships. Methods Psychol. 5, 100066 (2021).
12.
Labrecque J, Swanson SA. Understanding the assumptions underlying instrumental variable analyses: a brief review of falsification strategies and related tools. Curr. Epidemiol. Rep. 5, 214–220 (2018).
13.
Didelez V, Meng S, Sheeha NA. Assumptions of IV methods for observational epidemiology. Stat. Methods Med. Res. 19(4), 333–357 (2010).
14.
Greenland S. An introduction to instrumental variables for epidemiologists. Int. J. Epidemiol. 29(4), 722–729 (2000).
• Introduces instrumental variable (IV) methods to epidemiologists, explaining their assumptions, uses and limitations in observational research. Emphasizes the importance of careful interpretation and potential pitfalls in causal inference using IVs.
15.
Hernan MA, Robins JM. Instruments for causal inference: an epidemiologist's dream? Epidemiology 17(4), 360–372 (2006).
16.
Canan C, Lesko C, Lau B. Instrumental variable analyses and selection bias. Epidemiology 28(3), 396–398 (2017).
17.
Virolainen SJ, VonHandorf A, Viel K, Weirauch MT, Kottyan LC. Gene-environment interactions and their impact on human health. Genes Immun. 24, 1–11 (2023).
18.
Zheng J, Baird D, Borges MC et al. Recent developments in Mendelian Randomization Studies. Curr. Epidemiol. Rep. 4(4), 330–345 (2017).
19.
Haycock PC, Burgess S, Wade KH, Bowden J, Relton C, Davey Smith G. Best (but oft-forgotten) practices: the design, analysis, and interpretation of Mendelian randomization studies. Am. J. Clin. Nutr. 103(4), 965–978 (2016).
20.
Richmond RC, Davey Smith G. Mendelian randomization: concepts and scope. Cold Spring Harb. Perspect. Med. 12(5), a041302 (2022).
• Provides a comprehensive overview of Mendelian randomization (MR), outlining its conceptual foundations, assumptions and expanding scope in medical research.
21.
Sanderson E, Glymour MM, Holmes MV et al. Mendelian randomization. Nat. Rev. Methods Primers 2, 6 (2022).
22.
Jackson M, Marks L, May GHW, Wilson JB. The genetic basis of disease. Essays Biochem. 62(5), 643–723 (2018).
23.
Burgess S, Thompson SG. Collaboration CCG. Avoiding bias from weak instruments in Mendelian randomization studies. Int. J. Epidemiol. 40(3), 755–764 (2011).
• Examines the impact of weak instruments on bias in MR studies, demonstrating that low instrument strength can lead to estimates biased toward confounded observational associations.
24.
Burgess S, Small DS, Thompson SG. A review of instrumental variable estimators for Mendelian randomization. Stat. Methods Med. Res. 26(5), 2333–2355 (2017).
25.
Bhattacharya P. Linear regression in genetic association studies. PLoS ONE 8(1), e55008 (2013).
26.
Liu D, Gao X, Pan XF et al. The hepato-ovarian axis: genetic evidence for a causal association between non-alcoholic fatty liver disease and polycystic ovary syndrome. BMC Med. 21, 256 (2023).
27.
Palmer TM, Lawlor DA, Harbord RM et al. Using multiple genetic variants as instrumental variables for modifiable risk factors. Stat. Methods Med. Res. 21(3), 223–242 (2012).
28.
Lovegrove CE, Howles SA, Furniss D, Holmes MV. Causal inference in health and disease: a review of the principles and applications of Mendelian randomization. J. Bone Miner. Res. 39(8), 1539–1552 (2024).
29.
Angrist JD, Pischke JS. Instrumental variables in action: sometimes you get what you need. In: Mostly Harmless Econometrics: An Empiricist's Companion. Princeton University Press; 113–220 (2009).
30.
Palmer TM, Sterne JA, Harbord RM et al. Instrumental variable estimation of causal risk ratios and causal odds ratios in Mendelian randomization analyses. Am. J. Epidemiol. 173(12), 1392–1403 (2011).
31.
Cai B, Small DS, Have TR. Two-stage instrumental variable methods for estimating the causal odds ratio: analysis of bias. Stat. Med. 30(15), 1809–1824 (2011).
• Investigates the bias in two-stage IV methods for estimating causal odds ratios, highlighting that two-stage predictor substitution may introduce bias even without unmeasured confounding, while two-stage residual inclusion performs better but is still susceptible under certain condition.
32.
Schober PM, Vetter TR. Count data in medical research: poisson regression and negative binomial regression. Anesth. Analg. 132(3), 850–857 (2021).
33.
Payne EH, Gebregziabher M, Hardin JW, Ramakrishnan V, Egede LE. An empirical approach to determine a threshold for assessing overdispersion in Poisson and negative binomial models for count data. Commun. Stat. Simul. Comput. 47(6), 1722–1738 (2018).
34.
Garrod AB. Observations on certain pathological conditions of the blood and urine, in gout, rheumatism, and Bright's disease. Med. Chir. Trans. 31, 83–97 (1848).
35.
Dalbeth N, Choi HK, Joosten LAB et al. Gout. Nat. Rev. Methods Primers 1, 80 (2019).
36.
Choi HK, Atkinson K, Karlson EW, Willett W, Curhan G. Alcohol intake and risk of incident gout in men: a prospective study. Lancet 363(9416), 1277–1281 (2004).
37.
Zhang Y, Woods R, Chaisson CE et al. Alcohol consumption as a trigger of recurrent gout attacks. Am. J. Med. 119(9), 800–808 (2006).
38.
Choi HK, Atkinson K, Karlson EW, Willett W, Curhan G. Purine-rich foods, dairy and protein intake, and the risk of gout in men. N. Engl. J. Med. 350(11), 1093–1103 (2004).
39.
Choi HK, Curhan G. Beer, liquor, and wine consumption and serum uric acid level: the Third National Health and Nutrition Examination Survey. Arthritis Rheum. 51(6), 1023–1029 (2004).
40.
Chuah MH, Leask MP, Topless RK et al. Interaction of genetic variation at ADH1B and MLXIPL with alcohol consumption for elevated serum urate level and gout among people of European ethnicity. Arthritis Res. Ther. 26, 45 (2024).
41.
Rasheed H, Stamp LK, Dalbeth N, Merriman TR. Interaction of the GCKR and A1CF loci with alcohol consumption to influence the risk of gout. Arthritis Res. Ther. 19, 161 (2017).
42.
Sparso T, Andersen G, Nielsen T et al. The GCKR rs780094 polymorphism is associated with elevated fasting serum triacylglycerol, reduced fasting and OGTT-related insulinaemia, and reduced risk of type 2 diabetes. Diabetologia 51(1), 70–75 (2008).
43.
Onuma H, Tabara Y, Kawamoto R. The GCKR rs780094 polymorphism is associated with susceptibility of Type 2 diabetes, reduced fasting plasma glucose levels, increased triglycerides levels and lower HOMA-IR in Japanese population. J. Hum. Genet. 55(9), 600–604 (2010).
44.
Bound J, Jaeger DA, Baker RM. Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak. J. Am. Stat. Assoc. 90(430), 443–450 (1995).
45.
Sanderson E, Windmeijer F. A weak instrument [Formula: seetext]-test in linear IV models with multiple endogenous variables. J. Econom. 190(2), 212–221 (2016).
46.
Jalayer Khalilzadeh A, Tasci ADA. Large sample size, significance level, and the effect size: solutions to perils of using big data for academic research. Tourism Manag. 62, 89–96 (2017).
47.
Hall A, Rudebusch GD, Wilcox DW. Judging instrument relevance in instrumental variables estimation. Semant. Scholar 1996, 1–25 (1996).
48.
Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int. J. Epidemiol. 44(2), 512–525 (2015).
49.
Sanderson E, Glymour MM, Holmes MV et al. Mendelian randomization. Nat. Rev. Methods Primers 2, 6 (2022).
50.
Syed AAS, Fahira A, Yang Q et al. The relationship between alcohol consumption and gout: a Mendelian Randomization Study. Genes (Basel) 13(2), 321 (2022).
• Utilized MR to assess the causal relationship between alcohol consumption and gout, finding no significant evidence that genetically predicted alcohol intake increases the risk of gout or hyperuricemia.
51.
Nelson CR, Startz R. Some further results on the exact small sample properties of the instrumental variable estimator. Econometrica 58(4), 967–976 (1990).
52.
Bound J, Jaeger DA, Baker RM. Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak. J. Am. Stat. Assoc. 90(430), 443–450 (1995).
53.
Terza JV. Two-stage residual inclusion estimation in health services research and health economics. Health Serv. Res. 53(3), 1890–1899 (2018).
54.
Schuster NA, Twisk JWR, Ter Riet G, Heymans MW, Rijnhart JJM. Noncollapsibility and its role in quantifying confounding bias in logistic regression. BMC Med. Res. Methodol. 21, 136 (2021).
55.
Koladjo BF, Escolano S, Tubert-Bitter P. Instrumental variable analysis in the context of dichotomous outcome and exposure with a numerical experiment in pharmacoepidemiology. BMC Med. Res. Methodol. 18, 61 (2018).
56.
Zhang L, Lewsey J. Comparing the performance of two-stage residual inclusion methods when using physician's prescribing preference as an instrumental variable: unmeasured confounding and noncollapsibility. J. Comp. Eff. Res. 13, e230085 (2024).
• Assesses how unmeasured confounding and noncollapsibility affect the performance of two-stage residual inclusion methods using physician prescribing preferences as instruments. Highlights potential bias and variability in causal estimates under these conditions.
Information & Authors
Information
Published In
Copyright
© 2026 The authors. This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 Unported License
History
Received: 22 May 2025
Accepted: 13 February 2026
Published online: 18 March 2026
Keywords:
Topics
Authors
Metrics & Citations
Metrics
Article Usage
Article usage data only available from February 2023. Historical article usage data, showing the number of article downloads, is available upon request.
Citations
How to Cite
Comparison of two-stage methods for count data in Mendelian randomization: a simulation study. (2026) Journal of Comparative Effectiveness Research. DOI: 10.57264/cer-2025-0081
Export citation
Select the citation format you wish to export for this article or chapter.