Socioeconomic factors affect treatment delivery for patients with low grade glioma: a Swedish population-based study

We used multiple national registries to extract data on all adult patients (≥ 18 years) with a first-time diagnosis of supratentorial hemispheric diffuse LGG in Sweden in the period 2005–2015. Included tumors were WHO grade II astrocytoma, oligoastrocytoma or oligodendroglioma according to the 2000 and 2007 WHO classification of brain tumors [15, 16]. The SNOMED codes used were: 94003, 94203, 94113, 94103 for astrocytoma, 93823 for oligoastrocytoma and 94503 for oligodendroglioma (for oligoastrocytoma WHO grade II and oligoastrocytoma grade III the SNOMED code is the same 93823). Molecular tumor data was not available. Patients with radiologically suspected LGG but no histologically verified diagnosis were not included in the present study.

The patients were identified through the Swedish Brain Tumor Registry (SBTR), which is a regionally based registry of adult patients diagnosed with brain tumors covering data from 1999 and onwards. The SBTR contains detailed, prospectively collected, information on tumor and patient characteristics. The registration rate (defined as the percentage of diagnoses in the SBTR that corresponds to diagnoses reported to the compulsory National Cancer Registry) is generally high (> 90%) but has varied somewhat over time and between regions. In order to provide representative population-based data, we set a minimum registration rate of 80% as a requirement to be included in the analysis at any given year for each region. For this reason, only data from the period 2012–2013 were used in one (out of six) region, while only data from 2009 to 2015 were used in the case of another region. For the four remaining regions, data inclusion covered the entire period 2005–2015. For further details of the SBTR, see Asklund et al. [17].

Statistics Sweden is a government agency responsible for coordinating the system for the official and objective statistics for general information, investigation, and research in Sweden (www.​scb.​se). Using the unique personal identification numbers assigned to all Swedish residents, we were able to extract individual data on educational level and income for the patients in our cohort. Such individual level information is rarely available for research purposes and similar studies usually rely on area-based approximations for these factors [18]. We received data per year, using the factual income during the year prior to diagnosis, while educational level was registered for the year of diagnosis. The registry was accessed on June 26th 2017.

Numerous different but often interlinked indicators are used to measure socioeconomic position (SEP) in health research. In this study, we used income and education, where especially the latter is suggested to be a particularly rich indicator of both early life circumstances and adult resources, as well as occupational possibilities [19].

To obtain individual level information on comorbidity, we extracted information from the National Patient Registry (NPR), which is one of several registers governed by the National Board of Health and Welfare (NBHW). The reporting to NPR is mandatory, and we received data concerning inpatient and outpatient visits, including diagnostic and procedural codes in the period 2003–2016 excluding primary health care contacts. The underreporting of contacts in NPR has been estimated to be less than 1% according to the NBHW (www.​socialstyrelsen.​se). The ICD-10 codes were the basis for calculating the Elixhauser comorbidity index [20, 21]. The conditions removed from the index due to possible association with diagnosis of LGG were: G40; epilepsy, G41; status epilepticus, R56; convulsions, R47; dysphasia/aphasia and C70–72; Malignant tumor in central nervous system. Each patient received a score from 0 to 30 based upon comorbid categories present or not. We report categories as 0, 1, 2, and 3 or more.

All analyses were performed with the SPSS ver. 25.0 or newer software. The statistical significance level was set to p < 0.05. All tests are two-sided. Central tendencies are presented as means and standard deviations, or medians and interquartile ranges if skewed. Categorical data were analyzed with the Pearson Chi-square test. Comparisons of continuous variables between groups were analyzed using analysis of variance when normally distributed or the Kruskal–Wallis test if skewed. Time to event-analyses including overall survival are presented in Kaplan–Meier curves and compared using the log-rank test.

To calculate adjusted hazard ratios for univariable and multivariable effect on mortality and waiting time for surgery, we initially planned to use a Cox regression model. However, assumptions for proportional hazards were not met, which is why stratifications for relevant variables were used to create more homogeneous groups. After initial preplanned analyses on associations of income, educational level, and sex with clinical factors and outcome, additional post hoc analyses were made to assess the impact of comorbidity.

This study was approved by the Ethical Review Board in Västra Götaland region (Dnr 702–16).

Data on income level were available from 542 out of the 547 LGG patients. As shown in Table 1, patients in the lowest income group presented with worse pre-operative functional status compared to patients in the higher income groups. There were no significant differences between the groups regarding the waiting time for surgery, type of surgical procedure performed or number of complications leading to re-operation in relation to surgery. The post-hoc analyses on comorbidity revealed that more patients in the lowest income tertile had at least one comorbidity according to the Elixhauser comorbidity index (33.5%, N = 179) as compared to those in the highest income tertile (20.3%, N = 177, Table 1).

Information on educational level for 501 out of the 547 patients was available. We divided the patients into three commonly used categories of educational level, as outlined above. However, as seen in Table 2, this categorization yielded an uneven distribution between groups, with markedly fewer patients in the lowest educational level group (lower education N = 73, intermediate education N = 238, higher education N = 191).

Compared to the higher education groups, patients with the lowest educational level had a worse functional status upon presentation. Furthermore, this group waited longer from imaging diagnosis to surgery than patients with higher levels of education (Fig. 1). Finally, an increased comorbidity rate was found in the lowest educational group. No significant differences between the groups were found regarding type of surgical procedure or complications.

Similar to what was found for income and educational level, a difference in functional status at presentation was seen between women and men, with women being in a worse condition pre-operatively (Supplementary Table 1). In addition, a higher proportion of men compared to women were asymptomatic at presentation (9.9% versus 3.8%). There were no significant sex differences regarding waiting time for surgery (median 35 days for females and 36 days for males), type of surgical procedure performed, or number of complications leading to re-operation in relation to surgery.

When analyzing survival after surgery, we found that patients in the lowest income group and in the lowest educational group had shorter survival times compared to patients with higher income and education (Figs. 2, 3). No significant survival differences were seen between males and females.

To adjust for variables with group level differences found in earlier analyses and for factors likely to affect outcome, we made separate stratifications for age (including only patients aged 18–60), comorbidity (including only those without comorbidity) and pre-operative performance score (including only fully active patients).

When analyzing overall survival, the differences over educational levels and over income levels were no longer significant following age- and comorbidity stratification. However, for sex, a small survival difference to the benefit of females appeared in the stratum of fully active patients (log rank, p = 0.009).

Likewise stratified analyses, were made for waiting times to surgery with the addition of tumor size as a separate stratification (excluding tumors > 6 cm). All other analyses were stratified as outlined above. We decided beforehand on omitting extreme outliers (with more than three years from radiological diagnosis to surgery). The differences in waiting times were still associated with educational groups following these stratifications (age stratum log rank, p < 0.001, comorbidity stratum log rank, p = 0.001, size stratum log rank, p < 0.001, performance status stratum log rank, p = 0.03).

For groups defined by income, a significant difference in waiting times to the benefit of patients with high income level appeared in the stratum containing non-comorbid patients only (log rank, p = 0.006). No differences in waiting times between sexes were seen following the above-mentioned stratifications.

To evaluate whether potential differences in waiting time and educational level between different regions in Sweden underlay the disparities seen over social groups, descriptive sub-analyses were made for the different hospital regions (Supplementary Table 2). These analyses showed interregional differences in waiting time, as well as an uneven distribution of income levels and education. However, the longer waiting times for patients with low education compared that of patients with high education were present and consistently found within each separate region.

We found that female patients and patients with lower level of education or income presented with worse functional status. These findings imply that patients in lower socio-economic groups may reach specialized healthcare at a later stage of their disease than those with higher SEP. Impact of socioeconomy on stage of cancer at the time of diagnosis has been observed for other types of cancer, in both market-based health care systems such as in the US [4, 8, 18], and societies with universal health coverage [8, 22, 23].

Since the nature of our study is observational, we can only speculate as to the causes of the observed differences. The so-called patient’s delay as well as doctor’s delay may be influenced by the SEP of the patient. We also observed an association between low socioeconomic status and increased comorbidity, a finding that has been described earlier [24, 25]. It is possible that an increased complexity in interpreting symptoms of individuals with a high burden of comorbidity leads to a postponed diagnosis. This hypothesis is consistent with previous findings in patients with cervical cancer, where increased comorbidity was associated with a more advanced cancer stage at diagnosis [22].

Regarding sex, the male overrepresentation in head trauma [26, 27] may lead to a higher amount of incidental findings of brain tumors on trauma-related CT scans in men. This might also explain the better performance status at presentation on the group level for males in our cohort. In accordance, we found a higher proportion of asymptomatic males compared to females in our material (p < 0.01, Supplementary Table 1). Other possible explanations might be gender differences in care-seeking behavior and unequal response from the referring physician regarding symptoms presented by males or females. In the literature, examples of ‘gender gradients’ pertaining to delayed diagnosis or delayed access to specialist care can be found in both directions [8, 28‐30]. In our material, we found a slight overall survival benefit for females toward males in the subgroup of fully active patients without restrictions. It cannot be excluded that equally early diagnosing of tumors in both sexes could have the potential to further ameliorate the clinical outcome for women with LGG.

We observed longer waiting times for surgery for patients in the lowest educational level group, as compared to patients with higher levels of education. Corresponding differences were found only in subgroups (non-comorbid) for income level and not at all for sex. Inequality in relation to waiting time for surgery linked to educational level is particularly disturbing in a healthcare system that has the objective of providing equal care to the population. From our stratified analyses, we noted that the observed differences in waiting times do not seem to emanate from differences in co-morbidity, age or pre-operative performance status. Also, patients with larger tumors generally are likely to get surgery more promptly but among patients with small or medium-sized tumors, those with lower education still endured longer waiting times. To exclude that the observed disparities were only a consequence of geographical differences in educational level and waiting times, we made a descriptive sub-analysis exploring each hospital region separately. The findings of disparities over socioeconomic groups were consistent and confirmed also within the separate regions.

LGG are often slow-growing tumors, and there are cases in which the benefit of surgery is limited by the risks of the procedure. The decision as to whether and when to perform surgery is normally based on individual circumstances and made in consultation with the patient, even though growing evidence for early and extensive surgery has emerged over the last decades [31]. Importantly, as can be seen in Fig. 1, the differences in waiting times related to educational level are not evident during the early phase (first 10 weeks) after diagnosis. This suggests that for urgent or evident surgical cases, the level of education of the patient does not affect waiting times, whereas social factors are likely to impact patterns of care in more complex cases.

To summarize, our results show longer waiting times for a socially underprivileged group. There are some earlier published examples of increased waiting times for surgery related to lower educational level, which in our opinion strengthen the idea that our observations are valid and not random [32‐34].

Both low education and low income were associated with shorter survival time in unadjusted analyses, but these differences disappeared in adjusted analyses. Thus, although patterns of care do seem to differ, the overall survival was unaffected by level of education and income in our healthcare setting. This lack of difference in survival is not surprising as the median difference in waiting times between for instance lower and higher education was only 19 days, and there was no difference in type of surgical procedure or surgical complications. In addition to treatment factors, survival of LGG patients is largely determined by disease-related factors and there is no reason to believe that these parameters would be unevenly distributed over socioeconomic groups. Finally, survival may be a too insensitive outcome measure. Although the lack of difference in survival is obviously reassuring, it provides us with only limited evidence for arguing that LGG patients are handled equally.

Our study has several limitations related to the observational design and the limited level of clinical details provided by the registers. The categorization of patients into different income and educational levels was based on the year of diagnosis and the potential ‘evolution’ of these factors over time for an individual patient has not been analyzed.

Strengths include the population-based data acquired through the standardized, consecutive, and prospective reporting to the SBTR. Most studies on socioeconomic status and health rely on area-based estimations as proxies for individual data [18, 35]. Swedish public registries normally show extraordinary coverage and data quality [36, 37], enabling us to present results on individual level from the vast majority of patients operated for LGG in Sweden in the period 2005–2015.