Existing methods that use propensity scores for heterogeneous treatment effect estimation on non-experimental data do not readily extend to the case of more than two treatment options. In this work, we develop a new propensity score-based method for heterogeneous treatment effect estimation when there are three or more treatment options, and prove that it generates unbiased estimates. We demonstrate our method on a real patient registry of patients in Singapore with diabetic dyslipidemia. On this dataset, our method generates heterogeneous treatment recommendations for patients among three options: Statins, fibrates, and non-pharmacological treatment to control patients’ lipid ratios (total cholesterol divided by high-density lipoprotein level). In our numerical study, our proposed method generated more stable estimates compared to a benchmark method based on a multi-dimensional propensity score.
Get full access to this article
View all access options for this article.
References
1.
RosenbaumPRRubinDB. The central role of the propensity score in observational studies for causal effects. Biometrika1983; 70: 41–55.
2.
HiranoKImbensGW. Estimation of causal effects using propensity score weighting: An application to data on right heart catheterization. Health Serv Outc Res Method2001; 2: 259–278.
3.
FengPZhouX-HZouQ-M, et al. Generalized propensity score for estimating the average treatment effect of multiple treatments. Stat Med2012; 31: 681–697.
4.
BrownDWet al. A novel approach for propensity score matching and stratification for multiple treatments: Application to an electronic health record–derived study. Stat Med2020; 39: 2308–2323.
5.
FortinSPJohnstonSSSchuemieMJ. Applied comparison of large-scale propensity score matching and cardinality matching for causal inference in observational research. BMC Med Res Methodol2021; 21: 1–11.
6.
WilcockADet al. Reperfusion treatment and stroke outcomes in hospitals with telestroke capacity. JAMA Neurol2021; 78: 527–535.
7.
DehejiaRHWahbaS. Causal effects in nonexperimental studies: Reevaluating the evaluation of training programs. J Am Stat Assoc1999; 94: 1053–1062.
8.
ImaiKVan DykDA. Causal inference with general treatment regimes: Generalizing the propensity score. J Am Stat Assoc2004; 99: 854–866.
9.
AbadieAImbensGW. Large sample properties of matching estimators for average treatment effects. Econometrica2006; 74: 235–267.
10.
LechnerM. Identification and estimation of causal effects of multiple treatments under the conditional independence assumption. In: Lechner M, Pfeiffer F (eds) Econometric evaluation of labour market policies. Physica, Heidelberg, 2001, pp.43–58.
11.
ImbensGW. The role of the propensity score in estimating dose-response functions. Biometrika2000; 87: 706–710.
12.
YangSImbensGWCuiZ, et al. Propensity score matching and subclassification in observational studies with multi-level treatments. Biometrics2016; 72: 1055–1065.
13.
PowersSet al. Some methods for heterogeneous treatment effect estimation in high dimensions. Stat Med2018; 37: 1767–1787.
14.
DaiFDiaoMSingTF. Effects of rail transit on individual travel mode shares: A two-dimensional propensity score matching approach. Trans Res Part D: Trans Environ2020; 89: 102601.
15.
ZanuttoELuBHornikR. Using propensity score subclassification for multiple treatment doses to evaluate a national antidrug media campaign. J Educ Behav Stat2005; 30: 59–73.
16.
HanSRubinDB. Contrast specific propensity scores. Biostat Epidemiol2021; 5: 1–8.
17.
ChapinFS. Experimental Designs in Sociological Research. New York: Harper, 1947.
18.
CochranWG. Matching in analytical studies. Am J Public Health Nations Health1953; 43: 684–691.
19.
RubinDB. Matching to remove bias in observational studies. Biometrics1973; 29: 159–183.
20.
CochranWGRubinDB. Controlling bias in observational studies: A review. Sankhyā: The Indian Journal of Statistics, Series A1973; 35: 417–446.
21.
SmithAKarkJCasselJ, et al. Analysis of prospective epidemiologic studies by minimum distance case-control matching. Am J Epidemiol1977; 105: 567–574.
22.
RubinDB. Using multivariate matched sampling and regression adjustment to control bias in observational studies. J Am Stat Assoc1979; 74: 318–328.
23.
d‘AgostinoRB. Tutorial in biostatistics: propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med1998; 17: 2265–2281.
24.
HeckmanJJIchimuraHToddP. Matching as an econometric evaluation estimator. Rev Econ Stud1998; 65: 261–294.
25.
StuartEA. Matching methods for causal inference: A review and a look forward. Stat Sci: Rev J Inst Math Stat2010; 25: 1.
26.
FosterJCTaylorJMRubergSJ. Subgroup identification from randomized clinical trial data. Stat Med2011; 30: 2867–2880.
27.
LohW-YHeXManM. A regression tree approach to identifying subgroups with differential treatment effects. Stat Med2015; 34: 1818–1833.
28.
DengCet al. Practical guidance on modeling choices for the virtual twins method. J Biopharm Stat2023; 33: 653–676.
29.
SchwartzMHRiesAJGeorgiadisAG. Short-term causal effects of common treatments in ambulatory children and young adults with cerebral palsy: three machine learning estimates. Sci Rep2022; 12: 7818.
30.
HuLLinJ-YSigelK, et al. Estimating heterogeneous survival treatment effects of lung cancer screening approaches: A causal machine learning analysis. Ann Epidemiol2021; 62: 36–42.
31.
HuLJiJLiF. Estimating heterogeneous survival treatment effect in observational data using machine learning. Stat Med2021; 40: 4691–4713.
32.
RubinD. Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol1974; 66: 688.
33.
ImbensGWRubinDB. Causal inference in statistics, social, and biomedical sciences. New York, NY: Cambridge University Press, 2015.
34.
CaliendoMKopeinigS. Some practical guidance for the implementation of propensity score matching. J Econ Surv2008; 22: 31–72.
35.
StuartEAGreenKM. Using full matching to estimate causal effects in nonexperimental studies: examining the relationship between adolescent marijuana use and adult outcomes. Dev Psychol2008; 44: 395.
36.
BruscaSBet al. Role of critical care medicine training in the cardiovascular intensive care unit: survey responses from dual certified critical care cardiologists. J Am Heart Assoc2019; 8: e011721.
37.
RosenbaumPRRubinD. Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc1984; 79: 516–524.
38.
CochranWG. The effectiveness of adjustment by subclassification in removing bias in observational studies. Biometrics1968; 34: 295–313.
39.
SävjeFHigginsMJSekhonJS. Generalized full matching. Polit Anal2021; 29: 423–447.
BillingsleyP. Probability and measure. 3rd ed. New York, NY: John Wiley & Sons, 1995.
44.
HorvitzDGThompsonDJ. A generalization of sampling without replacement from a finite universe. J Am Stat Assoc1952; 47: 663–685.
45.
HiranoKImbensGWRidderG. Efficient estimation of average treatment effects using the estimated propensity score. Econometrica2003; 71: 1161–1189.
46.
AustinPCStuartEA. Moving towards best practice when using inverse probability of treatment weighting (iptw) using the propensity score to estimate causal treatment effects in observational studies. Stat Med2015; 34: 3661–3679.
47.
Ministry of Health of Singapore. Diabetes mellitus. MOH Clinical Practice Guidelines 2014.
48.
StoneNJet al. 2013 ACC/AHA guideline on the treatment of blood cholesterol to reduce atherosclerotic cardiovascular risk in adults: a report of the American college of cardiology/American heart association task force on practice guidelines. J Am Coll Cardiol2014; 63: 2889–2934.
49.
UK Prospective Diabetes Study Group. Efficacy of atenolol and captopril in reducing risk of macrovascular and microvascular complications in type 2 diabetes: Ukpds 39. BMJ: British Medical Journal 713–720 (1998).
50.
ColhounHMet al. Primary prevention of cardiovascular disease with atorvastatin in type 2 diabetes in the collaborative atorvastatin diabetes study (cards): Multicentre randomised placebo-controlled trial. Lancet2004; 364: 685–696.
51.
LeowMK-S. Characterization of the asian phenotype—an emerging paradigm with clinicopathological and human research implications. Int J Med Sci2017; 14: 639.
52.
National Healthcare Group, Singapore. Chronic Disease Management System (2015). Accessed: 2022-04-20.
53.
TohMLeongHLimBK. Development of a diabetes registry to improve quality of care in the national healthcare group in Singapore. Ann Acad Med Singap2009; 38: 546–546.
54.
GunapalPPGet al. Setting up a regional health system database for seamless population health management in Singapore. Proc Singapore Healthcare2016; 25: 27–34.
55.
AndersonKMOdellPMWilsonPW, et al. Cardiovascular disease risk profiles. Am Heart J1991; 12: 293–298.
56.
StevensRJKothariVAdlerAI, et al. The ukpds risk engine: a model for the risk of coronary heart disease in type ii diabetes (ukpds 56). Clin Sci2001; 101: 671–679.
57.
SnowVet al. Lipid control in the management of type 2 diabetes mellitus: a clinical practice guideline from the American college of physicians. Ann Intern Med2004; 140: 644–649.
58.
FriedewaldWTLevyRIFredricksonDS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem1972; 18: 499–502.
59.
LiFMorganKLZaslavskyAM. Balancing covariates via propensity score weighting. J Am Stat Assoc2018; 113: 390–400.
60.
ViscontiGZubizarretaJR. Handling limited overlap in observational studies with cardinality matching. Obs Stud2018; 4: 217–249.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
