12  Sensitivity

Compare

It is unclear how many proxies (k) should be selected to be included in the propensity score model as there is no clear guideline in the literature.

12.1 Sensitivity analysis for k

12.1.1 Create propensity score formula

Hence we iterate the process (change k parameter in get_prioritised_covariates function in step 4) and obtain odds ratio (exponentiation of log-OR) for each k. We varied k from 10 to 140.

Tip

OR estimates stabilizes around 1.5, shows variability below k = 50 and above 110

12.2 Sensitivity analysis for n

Tip

We varied n from 10 to 120, remaining everything else constant (e.g., k = 100).

Hence we iterate the process (change n parameter in get_candidate_covariates function step 2) and obtain odds ratio (exponentiation of log-OR) for each n. We varied n from 10 to 120.

Tip

OR estimates stabilizes around 1.5 for above n = 60.

Literature suggested that this restriction of n can be detrimental (Schuster, Pang, and Platt 2015). Hence in the original analysis we chose n such that that is larger than available empirical covariates.

Schuster, Tibor, Menglan Pang, and Robert W Platt. 2015. “On the Role of Marginal Confounder Prevalence–Implications for the High-Dimensional Propensity Score Algorithm.” Pharmacoepidemiology and Drug Safety 24 (9): 1004–7.