Inequalities in survival and care across social determinants of health in a cohort of advanced lung cancer patients in Quebec (Canada): A high‐resolution population‐level analysis

Abstract Background Advanced lung cancer patients exposed to breakthrough therapies like EGFR tyrosine kinase inhibitors (EGFR‐TKI) may experience social inequalities in survival, partly from differences in care. This study examined survival by neighborhood‐level socioeconomic and sociodemographic status, and geographical location of advanced lung cancer patients who received gefitinib, an EGFR‐TKI, as first‐line palliative treatment. Differences in the use and delay of EGFR‐TKI treatment were also examined. Methods Lung cancer patients receiving gefitinib from 2001 to 2019 were identified from Quebec's health administrative databases. Accounting for age and sex, estimates were obtained for the median survival time from treatment to death, the probability of receiving osimertinib as a second EGFR‐TKI, and the median time from biopsy to receiving first‐line gefitinib. Results Among 457 patients who received first‐line treatment with gefitinib, those living in the most materially deprived areas had the shortest median survival time (ratio, high vs. low deprivation: 0.69; 95% CI: 0.47–1.04). The probability of receiving osimertinib as a second EGFR‐TKI was highest for patients from immigrant‐dense areas (ratio, high vs. lowdensity: 1.95; 95% CI: 1.26–3.36) or from Montreal (ratio, other urban areas vs. Montreal: 0.39; 95% CI: 0.16–0.71). The median wait time for gefitinib was 1.27 times longer in regions with health centers peripheral to large centers in Quebec or Montreal in comparison to regions with university‐affiliated centers (95% CI: 1.09–1.54; n = 353). Conclusion This study shows that real‐world variations in survival and treatment exist among advanced lung cancer patients in the era of breakthrough therapies and that future research on inequalities should also focus on this population.


| INTRODUCTION
Lung cancer remains the deadliest cancer in Canada. 1 Roughly, half of all lung cancer patients are diagnosed at stage IV, when the 1-year survival probability is as little as 17%. 1,2 With the gradual introduction of breakthrough palliative treatments like precision medicine (i.e., targeted therapies in 2010 and immunotherapy in 2015), 3,4 survival trends are expected to rise in the overall and advanced (i.e., locally advanced or metastatic) lung cancer populations. 5,6 However, survival improvements in socioeconomically, sociodemographically, or geographically disadvantaged subpopulations may not necessarily occur at the same pace as in advantaged subpopulations. Despite a universal healthcare system providing free health services at the point of care for medically necessary treatments in all provinces, there is growing evidence of social inequalities in outcomes along the cancer control continuum in Canada. 7 Due to suboptimal care, survival from different cancers, including lung cancer, may be shorter in socioeconomically, sociodemographically, or geographically disadvantaged groups. [8][9][10][11] For example, cancer patients with low socioeconomic status (SES), immigrant status, or living in specific regions in Canada are less likely to receive oncology consultations, surgery, radiotherapy, chemotherapy, and supportive care near death. [12][13][14][15] Cancer treatment delays can also impact survival, 16 including delays for targeted treatment in advanced lung cancer. 17 However, the evidence for SES-and geography-based differences in timely treatment for lung cancer in Canada and elsewhere is mixed. 14,[18][19][20][21] There are very few lung cancer studies in Canada that focus on inequalities in survival by SES measures other than income or by sociodemographic status (SDS) (e.g., immigration), and on inequalities in palliative care with systemic treatments. 1,13 Also, income-based inequalities in survival are not evident for Canadian patients diagnosed with stage IV lung cancer in 2010-2011, 1 a period when precision medicine barely existed. Decision-makers in Quebec are increasingly interested in the real-world effectiveness of breakthrough palliative therapies for advanced lung cancer, like EGFR tyrosine kinase inhibitors (EGFR-TKI). While the survival of EGFR-TKI users in Quebec has been investigated, 22,23 inequalities in survival and treatments by patient subgroups remain unknown. Investigating these inequalities in Quebec is relevant, since Quebec accounts for 23% of the Canadian population 24 and about 16% of all Canadian immigrants in Canada. 25 Among the most populous provinces (Ontario, Quebec, British Columbia, and Alberta), it has the lowest median household income (Quebec vs. others in 2020 constant Canadian dollars: $53,800 vs. $63,000-$75,300) 26 and a higher percentage of individuals with educational attainment below upper secondary levels (Quebec vs. others: 12% vs. 7%-8%). 27 Studies of advanced lung cancer patients in the United States suggest that there may be SES-based inequalities in survival, 28 in the use of breakthrough palliative treatments, 29 and geography-based inequalities in the delay to initiate EGFR-TKI treatment. 30 This population-based retrospective study sought to fill a knowledge gap on survival inequalities and potential drivers of these inequalities across understudied characteristics of advanced lung cancer patients living in the current context of breakthrough therapies in Canada. Among patients who received the first-generation of EGFR-TKIs, gefitinib, as first-line palliative treatment in Quebec, we examined fluctuations in survival by income, education, material deprivation index, immigration, and geographic region, as well as variations in the use and delay of EGFR-TKI treatment (i.e., use of a second EGFR-TKI, osimertinib, and delay between the tumor biopsy and first EGFR-TKI). Results from this study may also be relevant to other countries with similar socio-economic contexts, such as developed or high-income countries (i.e., United States, United Kingdom, European Union, and Australia) that have social inequalities in lung cancer treatment and survival, 9,10,12,19,20 and countries with universal healthcare systems which include coverage of oral cancer medicines. 31

| Study design and data
We conducted a population-level retrospective study by selecting patients in Quebec whose first indication of lung cancer was between April 1, 2001 and March 31, 2019, and who received gefitinib as first-line palliative treatment in that period. Data came from health administrative databases managed by the Régie de l'assurance maladie du Québec (RAMQ), which capture information related to public insurance plans for health and prescription drugs, and by the Ministère de la Santé et des Services sociaux (MSSS), that capture inpatient and outpatient health care utilization by the Quebec population. Gefitinib is only reimbursed in Quebec to treat advanced lung cancer patients who are naïve to systemic palliative treatment. Focusing on this treatment thus allowed us to identify advanced lung cancer patients that are close in time to their diagnosis date.
The dates for the first indication of lung cancer and the initial lung cancer diagnosis were determined based on previously established methods. 32 Patients who filled a prescription for gefitinib at least once between April 1, 2001 and March 31, 2019 were identified in the community pharmacy services database through previously selected international nonproprietary names (INN) and drug identification numbers (DIN). 22,23 To form cohort 1 for survival analyses and study the use of osimertinib as a second EGFR-TKI, we identified the line of treatment associated with patients' first gefitinib treatment using a verified algorithm 22,23 and excluded all patients who did not receive gefitinib as a first-line palliative treatment. From this cohort (cohort 1), a separate cohort (cohort 2) was formed to study the time delay from receiving a tumor biopsy (or biopsy-related intervention) to receiving treatment. To form cohort 2, we excluded patients who did not receive a biopsy or received it more than 6 weeks before their lung cancer diagnosis and/or more than 6 months before their first gefitinib treatment. The date of death was captured with the health insurance registry (FIPA).
This study was part of a Health System Impact Fellowship of the Canadian Institutes of Health Research and was conducted at the Institut national d'excellence en santé et services sociaux (INESSS). De-identified data was accessible through a tripartite agreement between MSSS, RAMQ, and INESSS. 33 INESSS is not responsible for the content of this publication, however, the cohort used in this study is also included in a larger study by INESSS on EGFR-TKIs in Quebec. 22,23

| Outcomes
Survival time was calculated in months from the date of first gefitinib prescription to either the date of death or March 31, 2020: the administrative censoring time. Previously selected INNs and DINs were used to identify patients who received osimertinib as a second EGFR-TKI during their follow-up after receiving gefitinib. 22,23 Medical interventions involving biopsies that occurred between patients' diagnosis date 32 and their first gefitinib treatment were extracted from physician billing and hospitalization databases with expert-validated billing and intervention codes (Appendix A). 34,35 Treatment delays were calculated as the number of days between each patient's most recent biopsy and their first gefitinib treatment.

| Exposures and covariates
FIPA was consulted for patients' age, sex, and areas of residence (1: the census-based dissemination area [DA], which are the smallest standard geographic unit [400-700 persons] at the national level; and 2: one of 18 health regions) at the time of their first gefitinib treatment. Patients' DAs were linked to neighborhood-level SES and SDS, and geography variables developed by INESSS with the 2016 census data and the overall RAMQ population in that year. The DA-level SES and SDS variables included categorical information on median income after tax, percent of low education, and percent of individuals who immigrated to Canada. The census-based geography variable classified patients' DA based on urbanicity: (1) DAs in Census Metropolitan Areas (CMA), which include Montreal and other large cities, were urban; (2) DAs in Census Agglomerations (CA) were suburban; and (3) all other DAs (non-CMAs/CAs) were rural. Montreal was analyzed separately from other CMAs since it includes 86% of all immigrants in Quebec. 36 Patients DAs were also linked to Pampalon's material deprivation index, which was developed with Quebec's 2016 Census population and is a composite score of income, education, and employment measures at the DA-level that have been standardized to Quebec's age and sex distributions. 37 Another geography variable we used, that is health center-based, delineated patients' health regions according to the presence of university-affiliated medical centers and the distance of other health centers from large health services centers in Montreal and Quebec. 38,39 More details on each exposure are listed in Table B1 (Appendix B). Comorbidities were calculated with the Population Grouping Methodology, 40,41 which uses diagnostic codes to search for 226 health conditions (excluding lung cancer) in the 3 years leading up to the date of the first gefitinib treatment.

| Statistical analyses
Medians and proportions were calculated for patients' baseline characteristics. The median follow-up time in cohort 1 was estimated with the distribution of censoring times (reverse Kaplan-Meier method) with the prodlim package in R. 42 We used the survival package in R to conduct inverse-probability-of-treatment-weighted (IPTW) Kaplan-Meier analyses. 43 Stabilized weights for IPTW were calculated with predicted treatment probabilities obtained from multinomial logistic regressions. The weighing produced a pseudo-population with age and sex distributions at each category of SES, SDS, and geography that are like those in the overall unweighted population. For each exposure level, survival curves were plotted. The ratios of median overall survival (OS) times were also estimated from the Kaplan-Meier analyses, and 95% confidence intervals were obtained with a stratified case resampling method for right-censored data. 44 Osimertinib, a third-generation EGFR-TKI, can increase overall survival when prescribed in place of chemotherapy as a subsequent line after another EGFR-TKI. 45 We set out to identify variations in the use of this treatment, which might explain potential survival inequalities observed in the same cohort. Therefore, we ran age-and sex-adjusted quasi-Poisson regressions to estimate the ratios of marginal probabilities of receiving osimertinib as a second EGFR-TKI treatment across SES, SDS, and geography in cohort 1. Bootstrapping was used to obtain 95% confidence intervals. Osimertinib was first approved for coverage by Quebec's public drug insurance program in November 2018. We assumed that the fraction of patients in our study that died before this period, who could not receive osimertinib, was non-differential across SES, SDS, and geography. The validity of this assumption was tested in a secondary analysis that included only patients who were alive after October 31, 2018.
In cohort 2, we estimated the 25th, 50th, and 75th percentiles of treatment delays as the shorter, median, and longer delays, respectively, in each category of SES, SDS, and geography. The weighting method applied to cohort 1 was also used for this analysis. At each percentile (25th, 50th, or 75th), we estimated ratios of delays across categories of SES, SDS, and geography, and obtained bootstrapped 95% confidence intervals.
With an expected small sample, our exploratory analyses did not include statistical significance testing. We interpreted our results based on the direction and magnitude of the point estimates and the width of the associated confidence intervals.

| Cohort characteristics
Of the 552 patients who received gefitinib between April 1, 2001 and March 31, 2019, 457 patients (82.8%) received gefitinib as a first-line palliative treatment, so they were included in cohort 1. Of the 457 patients, 353 (77.2%) patients were included in cohort 2 ( Figure 1). Only 1 patient received their first gefitinib before the fiscal year 2011, and more than half of all patients received it in 2015 or afterward, both in cohort 1 23 and cohort 2. For patients in cohort 1, the median age at first gefitinib treatment was 70 years, and the median time to treatment from diagnosis was 3 months (IQR: 1.9-12.9 months) ( Table 1). A large percentage of patients were female (68.5%) or did not receive osimertinib as a second EGFR-TKI treatment (79.9%). Most patients had between 0 and 9 comorbidities (73%), of which half had less than 5 comorbidities (36.5%). A higher proportion of patients resided in areas with the lowest income quintile (26.0%), or the highest quintiles of low education (25.4%) or immigration density (31.9%). At least half of all patients resided in Montreal (55.6%) or in areas with university-affiliated health centers (51.0%). There were no major differences in cohort 2 in comparison to cohort 1, except for a slightly shorter time from diagnosis to F I G U R E 1 Flowchart of study population in each cohort. treatment (median: 2.6 months, IQR: 1.8-4.9 months). Because of the small sizes of the middle categories of income, education, and immigration variables, and the suburban category of the census-based region variable, we refrained from interpreting any comparisons made with these categories.

| Overall survival
With a total of 337 deaths, the median follow-up and overall survival times were, respectively, 45.3 months (IQR: 27.3-57.1) and 19.8 months (95% CI: 17.1-23.1 months). Our application of IPTW was successful in balancing age and sex distributions across categories of SES, SDS, and geographic location (Appendix C). Ageand sex-standardized survival patterns across SES, SDS, and geographic location in the weighted populations are presented in Figure 2. The ratios in median survival times in Table 2 that were obtained from the inverse probability-weighted survival curves show that patients from neighborhoods with the highest percentage of low education had a shorter survival time (high vs. low: 0.82; 95% CI: 0.64-1.17), as did those living in areas with the highest level of material deprivation index (high vs. low: 0.69; 95% CI: 0.47-1.04). In contrast, patients from neighborhoods with the highest immigrant density had a longer survival time (ratio, high vs. low: 1.23; 95% CI: 0.84-1.73). Patients living in urban areas other than T A B L E 1 Patient characteristics (unweighted populations). Montreal had a lower median survival time than those living in Montreal (ratio: 0.71; 95% CI: 0.52-1.04). The ratios of median survival times by income groups and health center-based regions were the smallest.

| Probability of receiving osimertinib as a second EGFR-TKI treatment
Estimates in Table 3

| Treatment delays
The overall median time between a biopsy and first gefitinib treatment was 39 days (IQR: . The largest ratios of treatment delays were observed by geographical location and across all quantiles of treatment delays in the weighted populations (

| DISCUSSION
Using population-based health-administrative data, we investigated whether inequalities in survival and treatment services exist among advanced lung cancer patients receiving gefitinib as first-line palliative treatment in Quebec. We observed shorter survival among patients residing in areas most representative of low education, high material deprivation, and minimal immigrant density in relation to the least representative areas. Patients from rural areas, and from urban areas other than Montreal, had similarly shorter median survival times than Montreal patients. Our conclusions are qualitatively like several studies in Canada and elsewhere that investigated inequalities in survival from lung cancer by income, immigration status, and rurality in stage non-specific populations. 14,28,46-53 Two Canadian studies report no associations of survival with income in advanced lung cancer patients or with rurality in lung cancer patients of whom at least 50% were diagnosed at stage IV. 1,50 However, these studies were conducted in a period where breakthrough treatments were not available or in a province with inequities in public coverage of oral versus intravenous cancer drugs (e.g., Ontario). 54 Percentage a (95% CI) Percentage ratio a (95% CI) The inequalities we observed in the probability of receiving osimertinib as a second EGFR-TKI were in the same direction as our survival inequalities. Given that, across material deprivation, we observed the largest variation in survival, but the smallest in receiving osimertinib, it is unlikely that longer survival solely explains the larger variations in osimertinib use observed across census-based regions, immigration, and education. In our secondary analysis, patients residing in intermediary, remote, or northern health regions had a considerably lower likelihood of receiving osimertinib in comparison to university regions (Appendix D), T A B L E 4 Delay in receiving first-line treatment after receiving a biopsy-related intervention (age and sex-standardized, weighted populations). confirming the possibility of confounding by calendar time in our main health region-based analyses. Our results on the use of an EGFR-TKI (i.e., osimertinib) across income are similar to a previous study in Ontario reporting a small positive relationship. 14 The inequalities in delay from the time of biopsy to first-line targeted treatment we observed were not in line with the survival inequalities. Montreal patients experienced longer delays but also longer survival than those from other census-based regions, and delays by health center-based regions were the largest despite lacking important survival differences. It is possible that there is a threshold effect and that the observed contrasts in treatment delays (e.g., 45 vs. 35 days) do not translate into meaningful survival differences. The overall median delay of 39 days we observed is in line with delays reported in other studies in Quebec: (1) a median of 36 days from biopsy to first targeted therapy 55 ; (2) a median of 29 days from diagnosis to first palliative treatment, 56 and; (3) ≤28 days from diagnosis to first targeted therapy for 66% of patients receiving it. 57 In our study, all medians of treatment delay were above 30 days, with the highest being 45 days for peripheral regions. While our estimates are close to targets adopted by Quebec institutions of 28 and 42 days from diagnosis to first lung cancer treatment, 55-57 the inter-regional variation highlights room for improvement. Institutions in Canada have also demonstrated the achievability of delays as low as 7-15 days from the first visit involving a pathological diagnosis of advanced lung cancer to targeted treatment initiation. 58,59 Since the 25th, 50th, and 75th percentiles of treatment delay in peripheral health regions were consistently 10 days longer than those in university regions, we believe that the health center-based gap in treatment delays is more likely due to health system factors than patients' disease severity.

SES and SDS variables
One limitation of this study is the small sample size which resulted in reduced statistical power and widening the confidence intervals. However, this is inevitable since approximately 12% of advanced lung cancer patients receive EGFR-TKIs in Quebec. 60 Our study was restricted to patients with public drug insurance, who may be older than privately insured patients. This may induce a selection bias given that access to private insurance is driven by income and employment, which are impacted by health and age, factors that also affect survival. However, we assume a low proportion of privately insured patients in Quebec among all gefitinib users during the study period. Patients aged 65 and over represent 70% of all new lung cancer cases 32,61 and belong to a provincial-level age group in which 90% of residents are registered for public drug coverage. 54,62 Our analysis also accounted for differences in age across the different income groups, which allowed us to partially control for potential selection bias pathways diluting the relationship between SES and survival. Even though Canada's universal public health care system does not include outpatient prescription drugs, Quebec has a universal drug insurance policy that requires all residents to have prescription drug coverage (residents without private coverage are covered by a public program). Furthermore, oral cancer drugs that are available on the public formulary are fully covered by the public program. 54 It would be interesting to repeat our analyses with health administrative data from other provinces in Canada that have a low potential for selection bias due to drug insurance. For example, Alberta Manitoba, and Saskatchewan, offer universal public coverage and full reimbursement of cancer drugs that are available on the public formulary, and have only 12%-13% of all take-home cancer drugs sales paid by private insurance. 63 Lastly, we estimated associations that cannot be explained by differences in age and sex; however, future analyses should jointly account for potential confounders like immigration (or geographical location), comorbidities, smoking, ethnicity, and calendar time.
To our knowledge, this is the first study investigating SES-, SDS-, and geography-based inequalities in survival and treatment in a cohort of advanced lung cancer patients in Canada, in the era of breakthrough therapies. Our analyses suggest that socioeconomic, sociodemographic, and geographic inequalities in survival and in care with EGFR-TKIs exist among patients receiving gefitinib as first-line palliative treatment. Given that the use of breakthrough therapies is expanding and that about half of all patients are diagnosed at an advanced stage, we expect to see survival inequalities over time, in a similar direction as in our study, in the advanced lung cancer population. Future studies on survival inequalities should therefore include this population in a causal framework to identify mediators that can be acted upon with policy and clinical interventions.

CONFLICT OF INTEREST STATEMENT
The authors declare no other conflicts of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the Institut national d'excellence en santé et services sociaux (INESSS). The data are not publicly available due to privacy and ethical restrictions.  (Table A1).    (Table B1).

APPENDIX C
Balance in distributions of age and sex before and after inverse probability weighting with propensity scores (Table C1 and Figure C1).
T A B L E C 1 Distributions of age and sex in cohort 1 before and after propensity score weighting, N = 457.

APPENDIX D
Secondary analysis on likelihood of receiving osimertinib in cohort 1 (Table D1).