Using ecological socioeconomic position (SEP) measures to deal with sample bias introduced by incomplete individual-level measures: inequalities in breast cancer stage at diagnosis as an example

Educational level and characteristics of invasive breast cancer: findings from a French prospective cohort

PURPOSE

Breast cancer (BC) characteristics are known to influence patients survival. Social differences have been reported by previous studies for those characteristics but questions persist because of inconsistent conclusions. We aimed to investigate the impact of education on BC stage, grade, and hormone receptor (HR) status, while adjusting for potential confounders including a broad set of health behaviors, anthropometric measures, and reproductive factors.

METHODS

In the French E3N cohort, 5236 women developed a primary invasive BC for which there was available information on stage, grade, and HR status. No multivariate analyses was performed for BC stage based on the lack of association in bivariate analyses. Odds ratios and confidence intervals were estimated using multinomial logistic regression models for BC grade or binomial logistic regression models for HR status of BC.

RESULTS

Women with a lower education were diagnosed with higher grade BC compared to women with a higher education (1.32 [1.12; 1.57]). This association was slightly attenuated after adjustment for covariates independently and more strongly affected in the fully adjusted model (1.20 [0.99; 1.45]). A significant association was observed between lower education and HR- status of BC (1.20 [1.02; 1.42]) attenuated after adjustment for age at first childbirth (1.12 [0.95; 1.33]).

CONCLUSION

In this cohort, education was associated with BC grade and HR status but not stage at diagnosis. The link between education and BC grade was not entirely explained by the different adjustments. A specific mechanism could be at play and deserves further investigations.

Inherent Bias in Electronic Health Records: A Scoping Review of Sources of Bias

Objectives

1.1Biases inherent in electronic health records (EHRs), and therefore in medical artificial intelligence (AI) models may significantly exacerbate health inequities and challenge the adoption of ethical and responsible AI in healthcare. Biases arise from multiple sources, some of which are not as documented in the literature. Biases are encoded in how the data has been collected and labeled, by implicit and unconscious biases of clinicians, or by the tools used for data processing. These biases and their encoding in healthcare records undermine the reliability of such data and bias clinical judgments and medical outcomes. Moreover, when healthcare records are used to build data-driven solutions, the biases are further exacerbated, resulting in systems that perpetuate biases and induce healthcare disparities. This literature scoping review aims to categorize the main sources of biases inherent in EHRs.

Methods

1.2We queried PubMed and Web of Science on January 19th, 2023, for peer-reviewed sources in English, published between 2016 and 2023, using the PRISMA approach to stepwise scoping of the literature. To select the papers that empirically analyze bias in EHR, from the initial yield of 430 papers, 27 duplicates were removed, and 403 studies were screened for eligibility. 196 articles were removed after the title and abstract screening, and 96 articles were excluded after the full-text review resulting in a final selection of 116 articles.

Results

1.3Systematic categorizations of diverse sources of bias are scarce in the literature, while the effects of separate studies are often convoluted and methodologically contestable. Our categorization of published empirical evidence identified the six main sources of bias: a) bias arising from past clinical trials; b) data-related biases arising from missing, incomplete information or poor labeling of data; human-related bias induced by c) implicit clinician bias, d) referral and admission bias; e) diagnosis or risk disparities bias and finally, (f) biases in machinery and algorithms.

Conclusions

1.4Machine learning and data-driven solutions can potentially transform healthcare delivery, but not without limitations. The core inputs in the systems (data and human factors) currently contain several sources of bias that are poorly documented and analyzed for remedies. The current evidence heavily focuses on data-related biases, while other sources are less often analyzed or anecdotal. However, these different sources of biases add to one another exponentially. Therefore, to understand the issues holistically we need to explore these diverse sources of bias. While racial biases in EHR have been often documented, other sources of biases have been less frequently investigated and documented (e.g. gender-related biases, sexual orientation discrimination, socially induced biases, and implicit, often unconscious, human-related cognitive biases). Moreover, some existing studies lack causal evidence, illustrating the different prevalences of disease across groups, which does not per se prove the causality. Our review shows that data-, human- and machine biases are prevalent in healthcare and they significantly impact healthcare outcomes and judgments and exacerbate disparities and differential treatment. Understanding how diverse biases affect AI systems and recommendations is critical. We suggest that researchers and medical personnel should develop safeguards and adopt data-driven solutions with a "bias-in-mind" approach. More empirical evidence is needed to tease out the effects of different sources of bias on health outcomes.

Socio-demographic inequalities in stage at diagnosis of lung cancer: A French population-based study

BACKGROUND

Diagnosing patients at a non-advanced stage has become a mainstay of lung cancer prevention and control strategies. Understanding socio-demographic inequalities in stage at diagnosis may improve the targeting of interventions on patients at higher risk. This study aimed to identify these socio-demographic determinants in a large-scale French population-based cancer registry.

METHODS

All incident lung cancers diagnosed between 2008 and 2019 identified from the Poitou-Charentes Cancer Registry (south-west France) were included. Stage at diagnosis was categorised as advanced/non-advanced (TNM III/IV vs I/II) according to the 8th TNM edition, the objective being to ensure a consistent level of prognosis over time. Socio-demographic variables included age, sex, the French European Deprivation Index (EDI) and patient's place of residence. Their impact on stage at diagnosis was quantified by multivariate logistic regression models with subgroup analyses by histological subtype.

RESULTS

Out of the 15,487 included patients, 75% were diagnosed at an advanced stage (66% to 95% depending on the histological subtype), 17% at a non-advanced stage and 10% at a non-specified stage. Multivariate analysis showed different patterns according to histological subtypes. In patients with adenocarcinoma, a higher risk of advanced stage was found for younger and older patients (u-shape), those most deprived, and those living in rural areas. The same effect of age was reported for squamous cell carcinomas, while no association was found for small-cell lung carcinomas.

CONCLUSIONS

This study highlighted substantial socio-demographic inequalities in stage at diagnosis, specifically for adenocarcinoma patients. Diagnosis strategies could be refined and strengthened in the non-smoker population, in which adenocarcinomas are mainly reported.

Socioeconomic and geographic disparities of breast cancer incidence according to stage at diagnosis in France

PURPOSE

Low socioeconomic background (SB) has been associated with lower breast cancer (BC) incidence and higher BC mortality. One explanation of this paradox is the higher frequency of advanced BC observed in deprived women. However, it is still unclear if SB affects similarly BC incidence. This study investigated the link between SB and early/advanced BC incidence from Loire-Atlantique/Vendee Cancer registry data (France).

MATERIALS AND METHODS

Fourteen thousand three hundred fifty three women living in the geographic area covered by the registry and diagnosed with a primary BC in 2008-2015 were included. SB was approached by a combination of two ecological indexes (French European Deprivation Index and urban/rural residence place). Mixed effects logistic and Poisson regressions were used, respectively, to estimate the odds of advanced (stage ≥ II) BC and the ratio of incidence rates of early (stage 0-I) and advanced BC according to SB, overall and by age group (< 50, 50-74, ≥ 75).

RESULTS

Compared to women living in affluent-urban areas, women living in deprived-urban and deprived-rural areas had a higher proportion of advanced BC [respectively, OR = 1.11 (1.01-1.22), OR = 1.60 (1.25-2.06)] and lower overall (from - 6 to - 15%) and early (from - 9 to - 31%) BC incidences rates Advanced BC incidence rates were not influenced by SB. These patterns were similar in women under 75 years, especially in women living in deprived-rural areas. In the elderly, no association between SB and BC frequency/incidence rates by stage was found.

CONCLUSION

Although advanced BC was more frequent in women living in deprived and rural areas, SB did not influence advanced BC incidence. Therefore, differences observed in overall BC incidence according to SB were only due to higher incidence of early BC in affluent and urban areas. Future research should confirm these results in other French areas.

Influence of deprivation on hospital care pathways and costs for people with ischaemic stroke: A trajectory modelling study

BACKGROUND

Socioeconomic characteristics have a strong impact on the incidence of stroke. The highest levels of morbidity and mortality are associated with the lowest socioeconomic positions. Moreover, the health insurance budget for stroke is significant.

OBJECTIVES

To describe typical Hospital Care Pathways (HCPs) and their costs 1 year after acute stroke and to assess the impact of individuals' socioeconomic characteristics on HCPs and costs.

METHODS

This retrospective observational study used hospital data on people with primary ischaemic stroke in a region of France (Haute-Garonne). HCPs were identified by sequential analysis. Multinomial logistic regression was used to analyse the impact of socioeconomic characteristics on HCPs as measured using an ecological index of deprivation, and a gamma regression model was used to analyse costs.

RESULTS

The analysis identified 4 typical HCPs. The first HCP consisted exclusively of consultations, the second was composed of consultations and hospitalisations, the third included hospitalisations during the first 2 months after stroke followed by consultations, and the fourth included long-term hospitalisations. After adjustment, deprivation was associated with HCPs and determined its costs: a high level of social disadvantage determined the type of pathway and increased the costs.

CONCLUSIONS

This study is one of the first to assess the impact of social inequalities in health on post-stroke HCPs in France and their corresponding costs. Our results confirm the hypothesis that healthcare costs differ according to the level of deprivation.

Is survival rate lower after breast cancer in deprived women according to disease stage?

BACKGROUND

Socioeconomic deprivation has been associated with lower breast cancer (BC) survival, but the influence of stage at diagnosis on this association merits further study. Our aim was to investigate this association using the Loire-Atlantique/Vendee Cancer Registry (France).

METHODS

Twelve-thousand seven-hundred thirty-eight women living in the area covered by the registry and diagnosed with invasive breast carcinoma between 2008 and 2015 were included in the study. They were censored at maximal 6 years. Deprivation was measured by the French European Deprivation Index. Excess hazard and net survival were estimated for deprivation level, stage and age at diagnosis using a flexible excess mortality hazard model.

RESULTS

After adjustment by stage, women living in the most deprived areas had a borderline non-significant higher excess mortality hazard (+25% (95% CI: -3%; +62%)) compared to those living in the least deprived areas. Stage-adjusted 5-year net survival differed significantly between these two subgroups (respectively, 88.2% (95% CI:85.2%-90.5%) and 92.5% (95% CI:90.6%-93.9%)).

CONCLUSION

BC survival remained lower in deprived areas in France, despite universal access to cancer care. Intensification of prevention measures could help to reduce advanced BC, responsible for the majority of deaths from BC. A better understanding of remaining social disparities is crucial to implement specific interventions.

Assessing How Social Exposures Are Integrated in Exposome Research: A Scoping Review

BACKGROUND

Exposome research aims to describe and understand the extent to which all the exposures in human environments may affect our health over the lifetime. However, the way in which humans interact with their environment is socially patterned. Failing to account for social factors in research exploring the exposome may underestimate the magnitude of the effect of exposures or mask inequalities in the distribution of both exposures and outcomes.

OBJECTIVES

We aimed to describe the extent to which social factors appear in the exposome literature, the manner in which they are used in empirical analyses and statistical modeling, and the way in which they are considered in the overall scientific approach.

METHODS

We conducted a scoping review of the literature using three databases (PubMed, Embase, and Web of Science) up to January 2022. We grouped studies based on the way in which the social variables were used in the analyses and quantified the type and frequency of social variables mentioned in the articles. We also qualitatively described the scientific approach used by authors to integrate social variables.

RESULTS

We screened 1,001 records, and 73 studies were included in the analysis. Fifty-five (∼75%) used social variables as exposures or confounders or both, and a wide array of social variables were represented in the articles. Individual-level social variables were more often found, especially education and race/ethnicity, as well as neighborhood-level deprivation indices. Half of the studies used a hypothesis-free approach and the other half, a hypothesis-driven approach. However, in the latter group, of 35 studies, only 8 reported and discussed at least one possible social mechanism underlying the relationship observed between the social variable and the outcome.

DISCUSSION

Social factors in exposome research should be considered in a more systematic way, considering their role in structuring both the specific external and the internal exposome. Doing so could help to understand the mechanisms of construction and, potentially, alleviate social inequalities in health and mitigate the emergence of new ones. https://doi.org/10.1289/EHP11015.

The effect of social deprivation on the dynamic of SARS-CoV-2 infection in France: a population-based analysis

BACKGROUND

Data on health inequalities related to the dynamic of SARS-CoV-2 infection in France are scarce. The aim of this study was to analyse the association between an area-based deprivation indicator and SARS-CoV-2 incidence, positivity, and testing rates between May 2020 and April 2021.

METHODS

We analysed data reported to the Système d'Information de Dépistage Populationnel surveillance system between May 14, 2020 and April 29, 2021, which records the results of all SARS-CoV-2 tests in France. Residential addresses of tested individuals were geocoded to retrieve the associated aggregated units for the statistical information (IRIS) scale, corresponding to an area comprising 2000 inhabitants relatively homogenous in terms of socioeconomic characteristics. A social deprivation score was assigned to each area using the European Deprivation Index (EDI). We fitted negative binomial generalised additive models to model the age-standardised and sex-standardised ratios for SARS-CoV-2 incidence, positivity rates, and testing rates, and to estimate incidence rate ratios (IRRs) and 95% CIs of their association with EDI quintiles, using the first quintile (least deprived) as the reference category, adjusted for week, population density, and region.

FINDINGS

Analyses were based on 70 990 478 SARS-CoV-2 tests, of which 5 000 972 were positive. SARS-CoV-2 incidence was higher in the most deprived areas than the least deprived areas (IRR 1·148 [95% CI 1·138-1·158]) and positivity rates were also higher (IRR 1·283 [1·273-1·294]), whereas testing rates were lower in the most deprived areas than the least deprived areas (IRR 0·905 [0·904-0·907]). SARS-CoV-2 incidence and positivity rates remained higher in the most deprived areas than the least deprived areas during the second and third national lockdowns, and variation in testing rate was observed according to population density.

INTERPRETATION

Our results highlight a positive social gradient between deprivation and the risk of testing positive for SARS-CoV-2, with the highest risk among individuals living in the most deprived areas and a negative social gradient for testing rate. These findings might reflect structural barriers to health-care access in France and lower capacity of deprived populations to benefit from protective measures.

FUNDING

None.

Spatiotemporal ecological study of COVID-19 mortality in the city of São Paulo, Brazil: Shifting of the high mortality risk from areas with the best to those with the worst socio-economic conditions

BACKGROUND

Currently, Brazil is experiencing one of the fastest increasing coronavirus disease (COVID-19) mortality rates worldwide, with a minimum of 158,000 confirmed deaths presently. The city of São Paulo is particularly vulnerable because it is the most populated city in Brazil. Thus, this study aimed to analyse COVID-19 mortality in a spatiotemporal context in São Paulo, with respect to socio-economic levels.

METHOD

We modelled the deaths using spatiotemporal architectures and Poisson probability distributions using a latent Gaussian Bayesian model approach.

RESULTS

Both total deaths and confirmed deaths showed similar spatial patterns. Mortality was higher in men and increased with age. The most critical period regarding mortality occurred between the 20th and 23rd epidemiological weeks, followed by an apparent stabilisation of the epidemiological trend. The risk of death was greater in areas with the worst social conditions during the study period. However, this pattern was not uniform over time, since we identified a shift of high risk from the areas with the best socio-economic conditions to those with the worst conditions.

CONCLUSIONS

Our study corroborated the relationship between COVID-19 mortality and socio-economic conditions, revealing the importance of geographic screening in the integration of better actions to face the pandemic.

Importance of collecting data on socioeconomic determinants from the early stage of the COVID-19 outbreak onwards

Disadvantaged socioeconomic position (SEP) is widely associated with disease and mortality, and there is no reason to think this will not be the case for the newly emerged coronavirus disease 2019 (COVID-19) that has reached a pandemic level. Individuals with a more disadvantaged SEP are more likely to be affected by most of the known risk factors of COVID-19. SEP has been previously established as a potential determinant of infectious diseases in general. We hypothesise that SEP plays an important role in the COVID-19 pandemic either directly or indirectly via occupation, living conditions, health-related behaviours, presence of comorbidities and immune functioning. However, the influence of socioeconomic factors on COVID-19 transmission, severity and outcomes is not yet known and is subject to scrutiny and investigation. Here we briefly review the extent to which SEP has been considered as one of the potential risk factors of COVID-19. From 29 eligible studies that reported the characteristics of patients with COVID-19 and their potential risk factors, only one study reported the occupational position of patients with mild or severe disease. This brief overview of the literature highlights that important socioeconomic characteristics are being overlooked when data are collected. As COVID-19 spreads worldwide, it is crucial to collect and report data on socioeconomic determinants as well as race/ethnicity to identify high-risk populations. A systematic recording of socioeconomic characteristics of patients with COVID-19 will be beneficial to identify most vulnerable groups, to identify how SEP relates to COVID-19 and to develop equitable public health prevention measures, guidelines and interventions.