Article
Pre-specified matching analyses in an observational study to evaluate treatment effects in a rare disease
Search Medline for
Authors
-
Lena Herich
- Staburo, München, Afghanistan
-
Martin Proescholdt
- Klinik für Neurochirurgie, Universitätsklinik Regensburg, Regensburg, Germany
-
Hannes Buchner
- Staburo, München, Germany
| Published: | February 26, 2021 |
|---|
Outline
Text
Background: Matching algorithms, as the propensity score matching, were introduced a long time ago (see eg. Rosenbaum and Rubin [1]). However, they are nowadays not only used in epidemiological research but also more and more in clinical research [2]. Examples range from breakthrough therapies using historic controls [3], drug development in rare diseases [4] to post-randomisation events as for exposure response analyses [5]. Thus, more and more observational data are used within drug appraisals [6] but also to support health technology assessment analyses for payers after approval of drugs [7].
We present a case study of a retrospective observational study in a rare and difficult-to-treat brain cancer where an intrinsic randomization based on case by case decisions of the patient's health insurance contact person took place. A randomized controlled trial was not feasible in this specific indication and to improve the intrinsic randomization, if necessary, matching techniques were pre-specified before any data was released to the statisticians.
Methods: The optimal propensity score matching algorithm [8] with a 1-to-1 matching ratio based on 3 covariates without replacement was defined as main analysis. The main analysis was repeated a) with an additional covariate b) with the same covariates but with the greedy nearest neighbour algorithm with treated subjects being selected in a random order and c) with the same covariates, the greedy nearest neighbour algorithm and a caliper [9] of 0.25 of the pooled standard deviation of logit of the propensity score from the 2 groups, a widely utilized rule of thumb [1]. Quality of matching was evaluated based on the absolute standardized differences [3]. The R-package Matchit [10] with a pre-specified random seed was used. Furthermore, a naÏve analysis without matching and an analysis adjusting for the covariates used for matching were performed.
Results: All matching analyses reduced the absolute standardized differences in covariates between the two treatment arms to values below 0.25 [3] thereby verifying the necessity of matching. The results of the study endpoint analyses (Quality of life, progression free survival and overall survival) on the matched data changed with respect to those on original data in the expected direction. The main analysis showed opposed to retrospective cherry picking not the best results of all 6 pre-specified analyses.
The authors declare that they have no competing interests.
The authors declare that a positive ethics committee vote has been obtained.
References
- 1.
- Rosenbaum P, Rubin D. Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score. The American Statistician. 1985 Feb; 39(1):33–38.
- 2.
- Borah BJ, Moriarty JP, Crown WH, Doshi JA. Applications of propensity score methods in observational comparative effectiveness and safety research: where have we come and where should we go? J Comp Eff Res. 2014 Jan;3(1):63-78.
- 3.
- Friends of cancer research. Panel 1: Augmenting randomized confirmatory trials for breakthrough therapies with historical clinical trials data. Exploring whether a synthetic control arm can be derived from historical clinical trials that match baseline characteristics and overall survival outcome of a randomized control arm: Case study in non-small cell lung cancer. White paper. 2018.
- 4.
- Nony P, Kurbatova P, Bajard A, Malik S, Castellan C, Chabaud S, et al. CRESim; Epi-CRESim. A methodological framework for drug development in rare diseases. Orphanet J Rare Dis. 2014 Nov;18(9):164.
- 5.
- Yang J, Zhao H, Garnett C, Rahman A, Gobburu JV, Pierce W, et al. The combination of exposure response and case-control analyses in regulatory decision making. J Clin Pharmacol. 2013 Feb;53(2):160–166.
- 6.
- Garrison LP Jr, Neumann PJ, Erickson P, Marshall D, Mullins CD. Using Real-World Data for Coverage and Payment Decisions: The ISPOR Real-World Data Task Force Report. Value in Health. 2007 Sep-Oct;10(5):326-335.
- 7.
- Quigley, JM, Thompson JC, Halfpenny NJ, Scott DA. Critical appraisal of nonrandomized studies – A review of recommended and commonly used tools. J Eval Clin Pract. 2019 Feb;25:44 – 52.
- 8.
- Ho D, Imai K, King G, Stuart E. Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference. Political Analysis. 2007;15:199-236.
- 9.
- Austin PC. A comparison of 12 algorithms for matching on the propensity score. Stat ed. 2014 Mar; 33(6):1057-69.
- 10.
- Ho D, Imai K, King G, Stuart E. MatchIt: Nonparametric Preprocessing for Parametric Causal Inference. Journal of Statistical Software. 2011;42(8):1-28.
