Virtual Library
Start Your Search
Michael T Bretscher
Author of
-
+
P09 - Health Services Research/Health Economics - Real World Outcomes (ID 121)
- Event: WCLC 2020
- Type: Posters
- Track: Health Services Research/Health Economics
- Presentations: 2
- Moderators:
- Coordinates: 1/28/2021, 00:00 - 00:00, ePoster Hall
-
+
P09.58 - Replication of Overall Survival in Chemotherapy Arms of 1L NSCLC Trials using Real-World Data External Controls (ID 3008)
00:00 - 00:00 | Author(s): Michael T Bretscher
- Abstract
Introduction
Previous work suggests that replication of overall survival (OS) in control arms of non-small cell lung cancer (NSCLC) trials is possible using real-world data (RWD) external controls (EC). This study expands the previous work by using more recent NSCLC trials.
Methods
Data from four IMpower trials (150-wildtype [WT], 130-WT, 131, 132) were used. Patients with advanced NSCLC who met eligibility criteria were identified from the Flatiron Health EHR-derived de-identified database. Covariates were balanced using propensity scores via standardized mortality ratio weighting and assessed with standardized mean difference (SMD). OS was defined from time from treatment initiation (EC) or randomization (trial) to death, last contact (EC), or end of follow-up (trial). Cox regression was used to estimate hazard ratios (HRs) comparing RWD ECs to trial controls and to experimental arms. After exploring within IMpower150-WT, the final methodology was pre-specified for remaining trials.
Results
The sample size was smallest for nab-paclitaxel (130-WT), which is less common in routine clinical settings. Balance was achieved (SMD<0.1) for all covariates in all trials except for time from diagnosis to 1L initiation for 130-WT. RWD ECs were not statistically different from trial controls, although point estimates suggest worse OS for some RWD ECs (Figure 1). HRs for treatment efficacy were similar to original trial HRs, except for 130-WT and 131 (Figure 2). Percent of patients treated with checkpoint inhibitors (CPIs) in any subsequent line ranged from 54.5% to 57.5% in RWD ECs and 19.3% to 46.4% in trial controls versus 2.8% to 10.3% in experimental arms. Sensitivity analyses using different trimming and weighting methods did not substantially change these results.
Conclusion
RWD ECs replicated OS well in two trials (150-WT, 132). Small sample size, use of CPIs in subsequent lines, and other unmeasured clinical characteristics may contribute to observed differences.
-
+
P09.60 - Using Patient-Level Data from Electronic Health Records to Replicate PFS Outcomes for Four 1L NSCLC Trials (ID 3009)
00:00 - 00:00 | Presenting Author(s): Michael T Bretscher
- Abstract
Introduction
While previous reports have demonstrated that external controls (EC) derived from electronic health record (EHR) data replicate overall survival (OS) of randomized clinical trial (RCT) controls, there is a need to evaluate non-OS endpoints. This study evaluates consistency of outcomes between progression-free survival (PFS) in external real-world controls (rwPFS) and RECIST-based investigator-assessed PFS from RCTs.
Methods
Data from four IMpower trials (150, 130, 131, 132) were used. Patients meeting trial criteria and receiving the respective comparator treatment were selected as ECs from the US-based Flatiron Health EHR-derived de-identified database. Confounders (age, sex, race, stage, smoking history, time since metastatic diagnosis) were adjusted for using standardized mortality ratio (SMR) weighting. The rwPFS endpoint definition included all progression events (including those without radiographic confirmation and events shortly after baseline), follow-up for progression beyond change of treatment, and deaths <30 days after end of progression follow-up. Methodology was pre-specified, allowing for modifications after exploration of IMpower150 (particularly of rwPFS variants) and subsequent application to the remaining trials. Cox regression was used to estimate hazard ratios (HRs) comparing real-world data ECs to RCT controls and to RCT experimental arms.
Results
Baseline demographics and clinical characteristics were balanced between RCT and real-world control arms except for time from diagnosis to 1L start date in IMpower130. When individual trial control arms were replaced with real-world data, rwPFS had similar results to RCT PFS results (Figure 1). Direct comparison of ECs vs RCT control arms resulted in HR’s close to unity (HR range: 0.88-1.07; Figure 2).
Conclusion
rwPFS outcomes using real-world controls were similar to RCT PFS results, supporting rwPFS as a potential endpoint for use with real-world-derived ECs. Limitations include insufficiently known patient or clinic level factors such as irregular timing of assessments, lack of health insurance, other physician involvement or not using RECIST criteria.
