Introduction
Head and neck cancer (HNC) accounts for approximately 3% of all new malignancies in Germany, and is ranked the seventh most common cancer worldwide (Global Burden of Disease Cancer et al. ,
2017). While the effect of socioeconomic factors (SES) on HNC survival has been documented in past literature (Boing et al.
2011; Choi et al.
2016; Johnson et al.
2008), recent studies have started to investigate the effect of area-based socioeconomic deprivation on cancer survival in general (Chang et al.
2012; Rachet et al.
2010; Singh and Jemal
2017), and HNC in particular (Bryere et al.
2017; Chang et al.
2013; Hagedoorn et al.
2016; Megwalu
2017). In Germany, however, studies investigating socioeconomic disparity are scarce and are often limited to certain regions (Brenner et al.
1991; Eberle et al.
2010; Finke et al.
2020; Jansen et al.
2020; Kuznetsov et al.
2011). Jansen et al. published the only large-scale study from Germany that aimed to measure social inequalities in cancer survival in 2014 (Jansen et al.
2014). This study found the 5‐year age‐standardized relative survival of the most deprived patients diagnosed with cancer of the mouth/pharynx to be 45.2% versus 49.3% for the most affluent patients. It is therefore essential to understand the mechanism by which social disparity affects cancer survival and to identify modifiable risk factors.
In this study, we aimed to (1) measure the survival gap according to socioeconomic deprivation level and (2) to decompose the total effect of deprivation on HNC survival into direct effect and indirect effect mediated through other possible factors. To this end, we used population-based and routinely collected data for patients diagnosed with HNC within Germany.
Indirect effect: role of deprivation and mediators
During the first 6 months after diagnosis, stage at diagnosis seemed to mediate most of the effect of deprivation across the more deprived quintiles. Using a counterfactual reasoning, the odds of dying of the patients in the most affluent quintile would increase by 44% ([OR] 1.44, 95% CI 1.32–1.58) if they were to be diagnosed as patients in quintile five (while keeping their level of deprivation, medical care, and treatment received unchanged and adjusting for age, sex, and year of diagnosis).
One year after diagnosis, the mediated effect of differential stage at diagnosis is only apparent in the fourth and fifth quintile. As follow-up time increases, there was no evidence that the considered mediators could contribute to the effect of deprivation on survival. Medical care and differential treatment seem to play no relevant role in mediating the effect of deprivation on survival (Table
3, Fig.
4).
Including tumor site as a confounder or including imputed stage information did not alter our results (Appendix 1, 2).
Discussion
Patients living in the most deprived districts at the time of diagnosis, showed the lowest survival rates according to our analysis. The total effect of deprivation seemed to be strongest during the first six months after diagnosis. While the effect subsided considerably at later time points, the survival disparity between the most deprived and most affluent remained substantial after 5 years. Our mediation analysis showed that stage at diagnosis played a major role in mediating the effect of deprivation within the first 6 months after diagnosis. Its role diminishes, however, as follow-up time increases. In contrast, there was no evidence that treatment and medical care mediated any of the effect of deprivation on survival throughout the study period.
Given that our study is based on a large sample size drawn from the national cancer registry, our results confirmed the survival disparity between the deprived and affluent patients in Germany, which is in line with Jansen et al. (Jansen et al.
2014). This survival gap, however, is difficult to explain in light of the universal health care system present.
To our knowledge, this is the first study that employs a counterfactual causal inference approach to gain a comprehensive understanding of the direct and mediated effect of social disparity on HNC survival in Germany. Through our DAG, we presented a detailed framework to analyze causal relations and to identify potential factors that could help explain the effect of socioeconomic deprivation. By having a clear visualization of the causal relations among variables, we were able to avoid potential biases (such as indication bias or selection bias), which could arise, for example, from the medical care-comorbidities-treatment relationship.
Based on the current literature available, we presented three potential mediators: medical care, stage at diagnosis, and treatment. Medical care for instance, was included as a mediator in our analysis based on the inequalities in health care utilization and availability experienced in Germany (Geyer
2008; Klein and von dem Knesebeck
2016). Patients from lower socioeconomic groups have been found to visit specialist practitioners less frequently, when compared with groups that are more affluent (Gruber
2010). Furthermore, results from a systematic review by Klein et al., suggested that major inequalities result primarily from prevention strategies, such as cancer screening (Klein et al.
2013).
Remarkably, in a study that investigated the effect of deprivation on breast cancer survival, Li et al. found that 35% (23–48%) of the higher mortality experienced by most deprived patients at six months after breast cancer diagnosis, was mediated by adverse stage distribution (Li et al.
2016). While stage at diagnosis is already recognized as a major prognostic factor in cancer survival, these results are interesting considering the wide availability of an advanced health care system in the UK, which is comparable to Germany.
Medical care along with minor vs. advanced treatment, on the other hand, revealed no evidence in mediating the effect of deprivation. Since the standardized “quality of health care” index is not available on a district level, we included the number of hospital beds (in the three previously mentioned departments) per districts’ population as an indicator of health-care availability and access. Information, like health insurance coverage status (private vs public) or waiting times however, were not available in our measurement. In a study by Lungen et al., patients covered by the statutory health insurance (public option) were found to wait 3.08 times longer for an appointment than private health insurees in Germany (Lungen et al.
2008). Lacking this information could have led to the underestimation of the mediated effects of these factors. Moreover, missing-stage information could have also played a significant role in this regard. A large proportion of missing treatment information (49.3%) was linked to patients living in the most affluent districts (Appendix 2). This was confirmed by our stepwise logistic regression that revealed deprivation level, age, medical care, and stage as the most significantly associated variables to missing treatment information (Appendix 2). In contrast, only a small percentage of stage information was missing (4.6%).
From a clinical perspective, it seems surprising that treatment fails to mediate the mentioned effects. This could be explained by that treatment cannot compensate for the adverse survival prospect due to an advanced stage. However, in our analysis, we could not fully account for details of the treatment, such as the intent of treatment, administered radiation dose, the chemotherapy given, or the surgical procedure performed. Treatment in the form defined seems to be universally available and might follow the average health performance in a district that determines the received treatment.
Considering that the development of HNC is a multifactorial process associated with a variety of risk factors, we have also presented an alternate route in our DAG that could also explain the effect of deprivation on survival. Major risk factors that were missing in our dataset, such as tobacco, alcohol consumption, and comorbidities have already been established as prognostic variables that are directly influenced by socioeconomic factors. In addition, HPV infections have been recently linked to up to 25% of HNC cases (Kreimer et al.
2005). Patients diagnosed with HPV-positive HNC were more likely to be younger men, non-smokers, and have higher SES when compared with HPV-negative HNC patients (Gillison et al.
2008). HPV-positive oropharyngeal carcinoma is also associated with better response to treatment and better survival (Ang et al.
2010; O’Rorke et al.
2012). It was, therefore, necessary to address potential bias that might arise from the missing HPV status. The pathologic evaluation of HPV status is currently based on PCR-based strategies, type-specific in situ hybridization(ISH) techniques, and immune-histochemical detection of surrogate biomarkers (e.g. p16 protein) (Westra
2009). Tumors positive both for p16 immunochemistry and HPV ISH are usually classified as HPV-positive (Robinson et al.
2010). While acknowledging this as a limitation, we performed our sensitivity analysis based on tumor site, which we considered a proxy for the missing HPV status. We found no significant differences in tumor-site proportions according to deprivation, nor did our results change when we included tumor site as an additional confounder in our Cox regression and mediation analysis.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.