The incidence risk of type 2 diabetes mellitus in female nurses: a nationwide matched cohort study

A retrospective and longitudinal method was adopted in this study, and data were sourced from the National Health Insurance Research Database. To comply with privacy measures, personal information was removed from the collected data.

The Taiwanese government created a compulsory National Health Insurance (NHI) system in 1995, a governmental insurance system that all citizens are mandated to be insured under. Since the end of 2012, the NHI coverage rate reached to as high as 99.85 % [14]. The NHI covers costs for outpatient procedures; emergency procedures; and inpatient prescriptions, treatments, and examinations. The NHI database details all medical claims records for the insured, including treatments for chronic illnesses such as diabetes [15, 16].

Data were sourced from the Longitudinal Health Insurance Research Database of 2000 published by the NHI, which included one million population randomly selected to be representative of whole population in Taiwan and a sample of nurses was selected from the nationwide Registry for Medical Personnel published by the NHI. We used female nurses and female non-nurses as the study participants. Nurses were defined as the nurses were listed in the Registry for Medical Personnel before December 31, 2000. Non-nurses were defined as the participants who had not registered as a licensed medical professional, such as physicians, dentists, physical therapists, nutritionists, and so on before the end of study observation. Participants younger than 20 years and older than 90 years and those diagnosed with type 2 DM prior to December 31, 2000 were excluded. Male nurses were excluded from the study, as most nurses in Taiwan are female (98.92 %) [17]. A total of 374,173 participants were selected, 70,675 of whom were nurses and 303,498 of whom were female non-nurses.

To objectively compare diabetes development risks among female nurses and female non-nurses, 1:1 propensity score matching was used to control for selection bias, and participants with similar characteristics and health conditions were selected from the two groups. After propensity score matching, we assigned 55,835 nurses to the observation group and 55,835 non-nurses to the control group (Table 1). Patterns of diabetes development were monitored until December 31, 2009, covering an average of 9.68 ± 1.06 years (9.72 ± 0.92 and 9.64 ± 1.19 y for nurses and non-nurses, respectively).

In this study, the type 2 diabetes mellitus was defined as a primary or secondary diagnosis with ICD-9-CM: 250 or A-code: A181 and the patients had made 3 or more clinic visits or been hospitalized at least once within 365 days [18], and this study excluded other types DM patients. We categorized residence areas into 7 levels, and level 1 was the most urbanized [19]. We used the modified Charlson Comorbidity Index (CCI) to classify the severity of comorbidity [20]. The CCI involved 17 comorbidities weighted based on severity. Higher score denoted greater comorbidity. Presence of catastrophic illnesses was defined as yes or no. The catastrophic illnesses or injuries were defined by National Health Insurance Administration in Taiwan including 30 categories of major illnesses (e.g., cancer, stroke, hemophilia, type I diabetes, autoimmune diseases, end-stage renal disease etc.) for which patients were exempted from copayment and thus avoided financial hardship [21].

The data prior to propensity matching revealed significant differences (p <0.05) between age, monthly salary, and CCI levels for the two groups (Table 1). Before propensity score matching, the average age of the nurses was 31.23 ± 8.48 years, whereas that of the non-nurses was younger by 10.97 ± 7.00 years at 42.20 ± 15.48 years. Most of the nurses (34.35 %) earned a salary of NT$ 36,301–NT$ 45,800, and the non-nurses earned NT$ 17,281–NT$ 22,800 (44.25 %) on average. The CCI scores for the nurses and non-nurses were 2.93 ± 2.89 and 3.48 ± 3.43, respectively.

After performing the propensity score matching tasks, we did not find any significant differences in terms of age, monthly salaries, residence area urbanization levels, CCI levels, and in other catastrophic illness levels between the two groups (Table 1).

In the Table 2, the results of the bivariate analysis revealed a lower incidence of diabetes among the nurses (2.68 %) compared with that of the non-nurses (3.13 %), revealing a statistically significant difference (p <0.05). After controlling for other factors, Cox proportional hazards models were used to identify the risk of diabetes for the nurses and non-nurses (Table 3). The results of the nurses showed a lower risk of developing diabetes compared with the non-nurses (Adj. HR = 0.84, 95 % CI: 0.79–0.90). For age, in the 35–44- and 45–54-year age groups, the nurses exhibited a significantly lower risk of diabetes compared with the non-nurses (Adj. HR: 0.82 and 0.75, respectively). For the income groups of less than NT$ 17,280, NT$ 22,801–NT$ 28,800, and NT$ 36,301–NT$ 45,800, nurses presented significantly lower risks of diabetes development compared with those of the non-nurses (Adj. HR: 0.65, 0.77, 0.86, respectively). Regarding urbanization variables, with the exception of participants residing in Levels 4 and 5 urbanization areas, no significant differences in diabetes development risk were found between the nurses and non-nurses (p <0.05). In less urbanized regions, nurses showed lower risks of diabetes development than non-nurses (p <0.05).

An analysis on the comorbidity of other catastrophic illnesses indicated that risks of diabetes development facing those nurses who did not have other catastrophic illnesses were significantly lower than those of non-nurses (Adj. HR: 0.84; p <0.05). While risks of diabetes development among the nurses with catastrophic illnesses were lower than those for the non-nurses, the results were non-significant (p <0.05). The results of the CCI analysis show that the nurses presented significantly lower susceptibility to diabetes development overall compared with the non-nurses (p <0.05).

The analysis results presented in Table 4 show that age and CCI serves as critical factors that influence diabetes development risks among nurses. Regarding age variables, a 10-year period was used as an interval to create age groups. The older nurses presented relatively higher risks of diabetes development compared with the nurses of less than 25 years (Adj. HR: 1.56–16.85). Nurses who were older than 65 years of age showed a 16.85-fold higher risk of developing diabetes compared with nurses who were younger than 25 years of age (95 % CI: 11.47–24.75).

CCI significantly influences diabetes development risks among nurses. The analysis results indicate that those nurses with a higher CCI were at a higher risk of developing diabetes than the reference group (CCI = 0). Those nurses with a CCI of ≧10 presented 28.48-fold higher relative risks of diabetes development than the reference group (95 % CI: 15.51–52.29). The research results shown in Table 5 indicate that diabetes incidence rates per 1,000 person-years for the nurses were lower than those of the non-nurses (2.76 ‰ vs. 3.25 ‰). The analysis results on the various variables reveal that with the exception of the ≦25-year age group, the incidence rate of diabetes per 1,000 person-years for nurses of all of the age groups was lower than that of the non-nurses.

There are several limitations to our analyses. According to the International Statistical Classification of Diseases (ICD), DM diagnosis is based on an ICD-9 diagnosis code. Therefore, clinical diagnostic data could not be acquired and verified. To optimize the accuracy of this study while compensating for this limitation, diabetes occurrence was identified when patients visited an outpatient department three times or when they were hospitalized more than once within 365 days of receiving a primary or secondary diabetes diagnosis (ICD-9-CM: 250 or A-code: A181) [17]. Some risk factors are not present in analysis. For instance, information on lifestyle, health behaviors and clinical testing data were not available and thus could not be used in variable analyses. So, we used the propensity score matching to control for selection bias. The propensity score adjustment is an important statistical technique to reduce the bias from confounding variables in observational studies and mimic the results of a randomized controlled trial [33]. In addition, the long follow-up and national design provided adequate power.