(103) Evaluation of Two Methods in Minimizing Confounding by Indication Bias in Observational Research: an Example Based on Guideline-recommended COVID-19 therapies

Background: The RECOVERY trial conducted in earlier stages of COVID-19 pandemic showed no difference and potentially increased mortality rates for dexamethasone vs. usual care among patients not requiring supplemental oxygen, i.e., low-severity COVID-19. Subsequent studies have shown a higher mortality rate with corticosteroids among patients with low-severity COVID-19 and clinical guidelines recommended against usage in low-severity population. When examining treatment effectiveness in observational research, appropriate methodology should be implemented to account for confounding and biases.

Objectives: Using two established methods to balance inherently different groups due to confounding by indication, we evaluated the effectiveness of remdesivir (RDV)+ dexamethasone (DEX) vs. DEX monotherapy among elderly patients hospitalized for COVID-19.

Methods: Building upon prior research on this topic, patients 65+ years hospitalised during the Omicron period (Dec’21-Apr’23) with a primary discharge diagnosis of COVID-19 flagged as “present-on-admission” who initiated RDV+DEX or DEX monotherapy within 2 days of hospitalisation were identified in the PINC AI Healthcare database. Patients were followed until day 28 or discharge status of expired/hospice, transfer, switching to another treatment group. Patients were categorized by baseline supplemental oxygen requirement: no supplemental oxygen charges, low-flow oxygen, high-flow oxygen/non-invasive ventilation, or invasive mechanical ventilation/ECMO. To achieve a balanced distribution of underlying confounders, two methods were used: 1) 1:1 propensity score matching (PSM) without replacement to estimate the average treatment effect on the treated (ATT) of RDV+DEX vs. DEX, wherein patients in the two treatment groups were matched within the specified caliper distance of PS. 2) Inverse probability of treatment weighting (IPTW), which estimates the average treatment effect (ATE) of RDV+DEX vs. DEX across the entire eligible cohort. Cox Proportional Hazards Model was used to assess time to 14- and 28-day mortality.

Results: In the study cohort, 41,228 initiated RDV+DEX and 25,579 initiated DEX monotherapy in the first 2 days of hospitalization.
Using PSM, 22,532 RDV+DEX patients were matched to 22,532 DEX monotherapy patients. Using PSM, the ATT estimates showed that RDV+DEX had a significantly lower mortality risk compared to DEX monotherapy across all supplemental oxygen requirements at 14-days and 28-days. Using IPTW, consistent results for the ATE were obtained.

Conclusions: Both ATT and ATE estimates obtained from PSM and IPTW respectively demonstrated consistent findings regarding the effectiveness of RDV+DEX vs. DEX monotherapy in reducing mortality among elderly patients.