A population-based study of the influence of socioeconomic status on prostate cancer diagnosis in Taiwan

All data were obtained from the following two population-based databases: the National Health Insurance Research Database (NHIRD) and the Taiwan Cancer Registry (TCR). These two databases can be linked by a unique encryption identity number. NHIRD was established in 1995 when the compulsory National Health Insurance (NHI) was first deployed [27]. By 2010, NHI had enrolled over 23 million citizens, representing 99% of the total population. According to NHI Act, Chapter 1 Article 2, every employed citizen must apply for the insurance by him/herself. The not-employed people can also join the insurance program through an employed close family member who must be a spouse or first linear ascendant. Thus, NHIRD defines beneficiaries by two insurance statuses: the employed insured and not-employed dependents, representing two different ways to join the NHI. Meanwhile, all beneficiaries can be classified into 15 social statuses. The insured are classified based on their occupational class, specific lifestyle or unique situation. Intriguingly, NHIRD assigns a reference status to dependents that is the same as their attached family member’s status.

The majority of beneficiaries were categorized as government employees, enterprise employees or members of the labor classes. The insured in the above occupations must be paid above minimum wage in conformance with legislative requirements, and the wage information was used to calculate individual insurance premiums. As the not-employed dependents had no wage, the wage of their attached family served as their reference income information. The other 12 statuses include employers, freelancers, soldiers, veterans, members of certain professional associations, monks and nuns, and sentenced persons. Each of the categories has its own characteristic features, and the incomes vary widely from category to category, ranging from the highest to below the poverty line. The not-employed individuals who also had no family associate to whom they could be attached were classified as a unique social status category. Accordingly, every beneficiary had a traceable social status and income information, i.e., his/her own or referenced from his/her family.

The second database, TCR, was founded by the Ministry of Health and Welfare in 1979. Hospitals with a capacity of more than 50 beds are obligated to report the detailed information of certain newly diagnosed cancers to TCR. PCa was not included in this mandatory reporting list until 2008 [28].

As a retrospective, follow-up, observational study, the study period was set from 2011 to 2013. The study populations were defined by the following eligibility criteria. All beneficiaries were first separated into two groups: the insured and the dependents. Next, we included male subjects who were actively covered on January 1, 2011 and who were aged between 50 to 79 years old. The age interval was determined based on the following considerations: (1) a large proportion of PCA cases are diagnosed in men aged more than 65 years; (2) PCa is an indolent and slow-growing disease, even though scholars have stressed that the greatest socioeconomic impacts on health occur in mid-adulthood, i.e., age 45–65 years [29]; (3) we referenced seminal researches on this topic and found that the age intervals were defined between 55 and 75 years [6, 7]; and (4) a lower incidence rate has been identified in Asian men, and the life expectancy of male Taiwanese in different counties is between 75 and 80 years.

Further, we retrieved subjects whose social statuses belonged to the categories of government employee, enterprise employee or labor class (Fig. 1), and we excluded people of the other 12 status categories. As these included subjects were all salaried employees, they must be paid above minimum wage (17,880 Taiwan Dollars (TWD) per month; approximately 600 USD). Subjects who had existing cancer diagnoses were also excluded (defined by The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes of 140.xx-208.xx). Moreover, PCa cases were collected from the TCR database. Cases with complete staging data were categorized into the early-stage group (stage I and II) or the late-stage group (stage III and IV) because of their distinct outcomes.

There were four independent variables in this study: (1) age – the most influential factor for PCa diagnosis based on current evidence; (2) beneficiary’s insurance statuses – either the employed insured or not-employed dependents; and (3 and 4) occupation and income – two common measures used to stratify SES. The study population was grouped based on occupational classes. Government employees included civil servants and teaching staffs. Most of these individuals held undergraduate degrees or above, and had to pass strict national examinations. In general, these subjects were better paid and had more confidence in their job security. Subjects in the labor class were farmers, fishermen and so-called blue-collar workers. As a rule, these individuals received less education and had lower income and less job security. The compositions of the enterprise employee category were more diverse than those of the other two groups, and the average education level, income, and job security of enterprise employee were expected to lie between those of labor class and government employee.

The statutory retirement age is 65 years, although the actual average retirement age is around 63 years for male citizens in Taiwan. However, the majority of labor workers continue to keep their jobs into their seventies. In general, Taiwanese government employees are recognized as more advantaged than enterprise employees, while the laborers are considered the least advantaged in the hierarchy.

As employers are required to report employees’ monthly wages, we decided to use this accurate and complete information to represent income status. In NHIRD, wage data are divided into fifty-four levels with ten different intervals. To avoid data manipulation, we used quantiles of the total population wage data to divide the study population into high-, middle- and low-income levels. Accordingly, high income was defined as a monthly wage of more than 40,100 TWD (the exchange rate was 30 TWD to one US Dollar); middle income as between 21,901 and 40,100 TWD; and low income as between 17,880 TWD (the minimum wage) and 21,900 TWD. As the income distribution was highly positively skewed, the subjects were concentrated in the low-income area, which resulted in more people in the low-income group than those in the other two groups (approximately 4:3:3).

In this paper, we used the insurance status of the beneficiary as the third categorical variable. As mentioned previously, the key difference between the beneficiaries was employment status, which has been recognized as a potential source of socioeconomic difference [30, 31]. We exploited this special arrangement to study whether there were health gradients within the not-employed population and, if there were, whether the patterns of health disparities of the dependents were the same as those of the insured.

We first applied a chi-square test to show the characteristic compositions of the insured group versus the dependent group. Next, we calculated the crude incidence rate and the percentage of late-stage diagnosis separately. After that, we subdivided each group into three age intervals and recalculated the ratios to show the age-adjusted effects.

Furthermore, we conducted four separate univariate analyses to test the individual influences of age, insurance status, occupation and income on PCa diagnoses. At the same time, we performed multivariable analyses in four models to illustrate their discriminating effects. In Model 1, we added the age variable to the insurance-status univariate analysis. Next, the occupation variable was added to Model 1 to create Model 2. For Model 3, we replaced the occupation variable of Model 2 with the income variable. In the full model, Model 4, all the independent variables were simultaneously analyzed. Using Cox proportional hazard model analysis, we analyzed the effects of the four variables on the PCa incidence rate. To find the influences of the independent variables on the odds ratio (OR) of late-stage presentation, we ran logistic regression analyses in the same sequence as described above.

We executed database cleaning, merging, and analysis using SAS, version 9.4 (SAS Institute Inc., Cary, NC). All analyses were two-tailed, and the level of significance was 0.05. All data were retrieved and processed within the Data Science Center of the Ministry of Health and Welfare, Taiwan.

In this study, we demonstrated that the crude incidence rate of PCa in 50 to 79 year-old males was 112/100,000 person-years (PYs), and approximately 44% of PCa cases were confirmed as late stage at diagnosis. Interestingly, the difference in the crude incidence rate between the insured and dependents was absent in the following three age interval comparisons (Fig. 2a). This absence was due to the combined influences of age on PCa diagnosis and the large differences in the age distribution between these two groups (Table 1). However, age did not show such dramatic effects on the proportion of cases with late-stage diagnosis. Nevertheless, a higher ratio of late-stage presentation in dependent group could be identified in Fig. 2b.

In the univariate analyses, all four variables demonstrated significant disparities in PCa diagnoses (all p < 0.01, in Table 2 and Table 3) except for the insurance status in the OR analysis (univariate model in Table 3). Unsurprisingly, age was evidenced to be the most significant risk factor of PCa [32]. The incidence rate of the 70 to 79 year-old group was around 12 times higher than that of the 50 to 59 year-old group (HR: 12.02, 95% CI: 11.22–12.89, p < 0.001), and the discriminating capability persisted throughout all analyses. Compared with the 50 to 59-year-old group, the 70 to 79 year-old group had a 70% higher tendency for late-stage presentation, and the 60 to 69 year-old group had a 24% higher tendency for late-stage presentation. The beneficiary’s insurance status predicted a false-positive significant difference in incidence rate, as explained above. There was no significant difference found in late-stage presentation between the insured and dependents. However, the unexpected trends in the intergroup health gradients demonstrated by the occupation and income indexes required further analysis.

Multivariable analyses showed that insurance status lost its explanatory power (all p > 0.05) in predicting disparities of PCa diagnoses except in Model 2 of Table 3 (p = 0.34) when analyzed together with age and occupation. Occupation and income revealed significant gradients in PCa diagnoses when analyzed with age and insurance status variables (Model 2 and Model 3 of Table 2 and Table 3).

In full model analyses, both occupation and income predicted that the advantaged subjects had higher incidence rates (all p < 0.05, Model 4 of Table 2). Unexpectedly, the middle-income group was found to have the lowest incidence rate (HR: 0.89, 95% CI: 0.82–0.96, p = 0.003), a finding that called for further elaboration. On the other hand, only the highest-income group demonstrated a significantly lower ratio of late-stage presentation (OR: 0.71, 95% CI: 0.58–0.85, p < 0.001). The other occupation and income indexes did not show significant intergroup disparities in late-stage presentation, as displayed in Model 4 of Table 3.

Our findings showed that, essentially, the more advantaged population had a higher PCa incidence rate and lower late-stage presentation ratio [22, 23, 25, 33]. However, the crude incidence rate of 112/100,000 PYs in our high-risk, senior study population was much lower than those found in Europe and the US [1, 2]. Similarly, an official report of the age-standardized PCa incidence rate, which was derived from the same databases in 2014, showed 29.1 per 100,000 PYs [34]. As to the staging statistics, in this population, 44% of PCa cases were late-stage presentations, while 56% showed an early stage at diagnosis, which is much lower than the current 80 to 92% rate of localized, early-stage diagnosis in the US [14, 35]. This information might be valuable for clinical professionals and policy makers, especially for those focusing on the Han Chinese population [36]. Additionally, these findings indicate that PCa is a critical and urgent public health challenge in Taiwan.

Some contextual information needs to be explained before interpreting and generalizing our results. The main characteristics of the health care system in Taiwan are that NHI is a mandatory, single-payer insurance system [37], and almost all hospitals have joined the system to serve this intensively competitive and highly efficient medical care market [27]. However, there has been no large-scale PCa screening trial in Taiwan. One possible explanation is that PCa was not one of the top five causes of male cancer until recently. When PCa is suspected, either incidentally through self-paid health examinations or symptomatically evidenced, patients only pay approximately 570 TWD (less than 20 USD) for a urological outpatient appointment and they can go directly to a tertiary medical center without a gatekeeper’s referral. Necessary diagnostic examinations and conventional therapies are comprehensively covered by NHI, while, unfortunately, no annotation had been made about why beneficiaries sought medical assistances in NHIRD. However, this does not mean that there is no barrier for the less-advantaged population to receive adequate medical care.

Occupation and income indexes are among the most commonly used measures in health disparity research [31, 38]. The basic assumptions suggest that higher incomes provide people with better nutrition, housing, schooling, and means to purchase health care resources. Also, more privileged occupations carry less risks of excessive job strain, occupational injury, exposures to toxic substances, and so forth. Both of which make the advantaged population healthier [38]. By and large, the advantaged are recognized as having a lower cancer incidence rate except for some frequently overdiagnosed cancers, such as PCa, breast cancer and thyroid cancer [39]. Researchers also identify stage at diagnosis as a complementary but important indicator to determine the phenomenon of overdiagnosis [40]. That is, when the advantaged suffer from a higher cancer incidence rate, they might simultaneously have the benefit of fewer late-stage presentations at diagnosis, if they are not experiencing complications resulting from overtreatment [41]. Consequently, the incidence rate combined with the stage at diagnosis must be examined together when performing cancer epidemiological research [24, 30, 39].

We did not find a significant difference in PCa diagnoses between the employed insured and not-employed dependents in this study. The main reason for this finding could be the choice of targeted disease. In addition, the health impacts of the dependent’s insurance status cannot fully represent the health effects of unemployment, which consist of voluntary and involuntary unemployed causes [42]. Indeed, at least half of our study population was older than the average retirement age, 62.8 years, for males in Taiwan (statutory retire age: 65 years), and there was no annotation for why dependents were not-employed. Furthermore, Shavers emphasized that the difficulties associated with classifying retired and unemployed subjects might lead to inconsistent results [31].

Notably, even though no differences were found between the insurance statuses in the incidence rate analysis, there was an exceptional finding of a statistically significant difference in late-stage diagnosis in Model 2 of Table 3 (p = 0.034, OR: 1.13, 95% CI: 1.01–1.27) and two borderline differences in Model 3 (p = 0.076) and Model 4 (p = 0.06). Although the differing effects were not remarkable in our study, we propose that the influence of the insurance status warrants further investigation, especially in younger populations and properly selected disease targets.

In this study, occupation showed significant discriminating capability in most of our analyses except in the full model (Model 4), as shown in Table 3. Hence, we conducted a Spearman’s correlation test of the four independent variables. The results highlighted that, on average, the advantaged occupation groups were younger and better paid, and a higher percentage of them were identified as dependents (correlation coefficient: − 0.122, 0.68, and 0.19, respectively; all p < 0.0001; n = 2,130,581). These correlations offered some explanation as to why occupation lost its differentiating capability, but the explanations were also outcome-indicator dependent. Our results support the arguments of Alder and Ostrove [43] stating that when more than one SES indicator was used, health outcomes might be correlated with one indicator more than the other.

Each group of researchers has its own preferences in measure selection, although most tend not to offer reasons for why their respective measures were chosen [17]. In Great Britain, occupation has been routinely collected in all official datasets and surveys [38] and was commonly used in seminal health inequality researches [44, 45]. While the status-list of NHIRD was originated from a mix of legacy professional insurance programs, it did not follow the occupation classification principles or standards of Western countries. Nevertheless, the individual occupational classes here represented an average and socially recognized level of education, income, job security and assessment of health care resources.

The majority of empirical findings support the assertion that more affluent populations have higher PCa incidence rates. Unexpectedly, we identified an enhanced intergroup disparity of the middle-income group, which showed the lowest incidence rate (HR = 0.89, 95% CI = 0.82–0.96, p < 0.003; Model 4 of Table 2). There are three probable explanations for this finding. First, epidemiological transition may have occurred, which means that the trends in socioeconomic disparities in health might have changed along with societal changes and economic developments [21, 46]. Second, disparities in PCa diagnoses is an exceptional case, especially when case growth had reached a plateau during study period. These suggestions could be verified by performing a similar study using other cancer types or through longitudinal approaches. Third, the result may be influenced by certain lifestyle factors among the local middle-income population that gives them special protection, such as green tea intake [47]. Although these hypotheses require further evaluation, our findings indicate that status syndrome [48, 49] may not apply to all scenarios.