Our findings highlight the improvement in survival of the admNSCLC patient population and suggest such improvements may be a result of changes in treatment over the past decade.
Our descriptive analyses build on recent work in broader populations of lung cancer and NSCLC patients [4, 9], demonstrating population level improvements in survival of subgroups of patients with admNSCLC over time which align well with the timing of the introduction of targeted and immunotherapies. However, by utilizing patient level EHR data on treatment and outcomes, we have been able to go further than previous studies, demonstrating that the changes in treatment patterns may plausibly explain the observed improvements in survival.
Significant improvements in survival among patients with non-oncogene positive tumors were observed from 2015 onwards, a timing that aligns with the earliest approvals of immunotherapies for this population. Adjustment for treatment in our main analysis provided an increase in the hazard of death by calendar year, suggesting that, independent of treatment, survival may have got worse over time. Adjusting for treatment in a more detailed manner in post-hoc analyses indicated that, independent of treatment, survival remained stable over time, suggesting that the categorization of treatment patterns in our main analysis may not have been granular enough to appropriately capture the relationship between treatment patterns, calendar time, and the outcomes. Notably, this post-hoc analysis suggests that changes in treatment pattern over time could explain the survival improvements observed.
Our main analysis that accounted for ALK inhibitors as a single class of drugs suggested treatment was unable to fully account for the improvements in survival observed among patients who had ALK+ tumors. The significant improvements in survival among patients with ALK+ tumors were observed between 2017 and 2019, a timing that aligns with the approval and introduction of second and third generation ALK inhibitors between 2017 and 2019. Our post-hoc analysis driven by this observation, in which we separated first and second/third generation ALK inhibitors into separate categories suggested that changes in treatment patterns over time could fully account for the survival improvements observed.
The absence of an improvement in survival among patients who had EGFR+ tumors is notable given we observed a large increase in the use of second and third generation EGFR inhibitors over our study period. Notably, the overall proportion of patients receiving any EGFR inhibitor (i.e., 1st/2nd/3rd gen) remained relatively stable over the study period suggesting that the incremental benefit of 3rd generation over 1st and 2nd generation products may not have translated into a large enough population level benefit over time for us to have the power to identify it in this study.
Patients with missing results for oncogene status were grouped into the non-oncogene positive cohort in our analyses; therefore, the impact of misclassification on our results must be considered. If these patients had ALK+ or EGFR+ tumors, their treatment patterns may have been inappropriately captured in our analyses thereby impacting the extent to which their survival over calendar time could be linked to treatment. Notably, findings remained relatively stable in sensitivity analyses investigating potential misclassification of patients with ALK+ and EGFR+ tumors into this population.
While we have been able to utilize patient level data to explore the relationship between treatment, calendar time, and outcomes, our data are observational in nature, and it is possible that other factors not fully captured in the main analysis may confound the relationships observed. Our sensitivity analyses including additional covariates provides some reassurance in this regard; however ,other changes that have occurred over the period in question may confound the associations observed; this might include broad changes in the healthcare system or in the standard of care of patients. Notably, mortality rate in the general population in the USA remained relatively constant over the period under study [10] while the classification of calendar time into annual periods overcomes additional challenges that can be encountered when comparing outcomes over calendar time, such as the impact of seasonality.
The small numbers of patients treated with some products led us to categorize treatments into classes and, in some cases, combine two classes of drug. While we have sought to group treatments in a logical manner, products within categories may have differing efficacy potentially introducing residual confounding. The short follow up available for patients diagnosed in later years may limit the power to detect differences in the patient populations with ALK+ and EGFR+ tumors given the relatively long survival in these groups. The post hoc nature of the analyses underlying a number of our key findings must also be considered. Finally, we classified patients into time constant treatment groups from baseline based on a pattern of treatments observed post-baseline. This may introduce immortal time biases, for example, owing to the fact patients receiving second line treatment had to survive long enough to receive such a treatment. In that regard, while we have provided some discussion linking the timing of improvements to the introduction of specific drugs, it is important to note that our study did not directly investigate the comparative effectiveness of different treatments or treatment classes and as, outlined above, the limitations of the study preclude one from drawing any such causal inferences from our results. Rather our study sought to explore whether changes in treatment over time may have been sufficient to explain the observed improvements in survival.
While these limitations need to be considered in interpreting our findings, we believe the fact we have demonstrated our findings across subgroups of the admNSCLC patient population, in which changes in the treatment paradigm have occurred at different points in time, suggests that changes in pharmacological treatment are as plausible an explanation for the findings as any of these other potential factors. As such, we believe our work provides a meaningful step beyond previous work in this area. Future work might seek to utilize data sources which can address a number of the above limitations or carry out similar analyses within a broader population of NSCLC patients which includes earlier stage patients.