The role of sickness absence diagnosis for the risk of future inpatient- or specialized outpatient care in a Swedish population-based twin sample

Reducing the extent of work incapacity in terms of sickness absence (SA) is highly prioritized in welfare countries. The incidence of SA is high and has also increased lately, even before the current pandemic of Covid-19, despite improved health conditions in European countries [1, 2]. Hence, absence from the labor market due to SA is a public health concern in several European countries [2]. A large amount of studies have previously reported associations between individual sociodemographic, socioeconomic, health- and work-related factors and future SA [3‐6]. However, far less studies have investigated the consequences of SA, although such are considerable both for the individual, employers and for society. For example SA is linked with a number of negative health-related consequences, such as disease (the same as being granted SA for, or another), worse well-being, weaker economy, less career development, worse social integration, and premature death [7]. Since SA is a mean to allow an individual to recover and retain work capacity, studies to investigate the consequences of various SA causes (i.e. diagnoses) have emerged.

SA is commonly prescribed in healthcare [8]. A lot of research have indicated that SA plays a role for many consequences, not only for disability pension (DP) [9, 10], but also for mortality or even suicide [11‐13]. However, even theoretically the effects of SA for an individual are complex and varying based on the life situation [14, 15], whereas SA is also known to increase the risk of reoccurring SA, DP and unemployment [16‐19]. On the other hand, besides SA, also patient care in terms of inpatient- or specialized outpatient care, is often needed for various symptoms and diseases [20, 21]. Some studies have addressed SA as a predictor of patient care [15, 22] although usually the pathway has been estimated utilizing the initiation from care, i.e. from onset of a disease, symptom or medication that required medical attention in a care unit, to SA [23, 24]. Therefore, a need exists to explore the consequences of SA at diagnosis group level to understand the process of SA [15].

Further, many chronic diseases, which also are grounds for SA, have moderate to large genetic influence. For example, genetics is known to play a role in low back pain [25], depression [26], anxiety [27], and e.g. blood pressure [28]. Since genetics is also known to affect SA [9, 22], there is a need for studies accounting for familial factors (genetics and mainly childhood environment) in associations of SA and potential consequences. A twin study with a co-twin control design utilizing twin pairs discordant for a factor of interest provides possibility to define if an association between SA and inpatient care is influenced by familial factors [29, 30].

In this study, we hypothesized that associations between SA due to a diagnostic group and patient care would follow broadly the severity of disease assumption [31] i.e. operationalized in this study by register data of diagnoses and inpatient vs. specialized outpatient care. Hence, we aimed to investigate if SA due to various diagnosis groups differ in the associations with subsequent inpatient- and specialized outpatient care. In specific, we aimed to control for the effect of familial confounding (genetics and early, shared environment) utilizing co-twin control design in the associations.

This study utilized the register data of the Swedish Twin project Of Disability pension and Sickness absence (STODS) [32] with all 119,907 twins from the Swedish Twin Registry (STR) born 1925–1990 [33]. One-third of the twins are monozygotic (MZ) and one-third same-sexed dizygotic (DZ) and last proportion opposite-sexed DZ twins [33]. For this study, the sample was limited to those ≥16 years of age, alive, living in Sweden, not on DP, or inpatient- or specialized outpatient care in 2005, and followed from the Micro-Data for Analyses of Social Insurance (MiDAS) database from the National Social Insurance Agency. We used the unique ten-digit Swedish identification number for the linkage of data from the national registers.

All residents in Sweden are eligible for the national SA system if they are ≥16 years of age, and have income from work, or unemployment or student benefits if they have a disease or injury causing work incapacity. All residents are also entitled to patient care [34].

Hence for the study period from 2005 until the end of 2013, we included dates and diagnoses for all SA (> 2 weeks) reimbursed by the Swedish Social Insurance Agency, in- and specialized outpatient care registries for dates and diagnoses from the National Board of Health and Welfare, and date of deaths from the causes of death register (for censoring). We could not control the SA before 2005. Emigration (date) was included as another censoring reason (from Statistics Sweden the Longitudinal Integration Database for Health Insurance and Labor Market Studies Register [LISA by Swedish acronym]) [35].

The final sample consisted of 69,552 twin individuals (Fig. 1). The number of complete twin pairs was 5416 MZ, 6757 DZ same-sexed, and 3251 DZ opposite-sexed twin pairs.

Those with first SA, emigration, or disability pension later than 2013 were excluded to avoid reversed causality, i.e. appearance of these shortly after the end of the follow-up. The final sample included censoring for International Classification of Diseases 10th Revision (ICD-10) codes O00-O99: Pregnancy, childbirth and the puerperium. Pregnancy and childbirth were not considered as illnesses in this study and since many might have SA or in- or outpatient care during that time, we did not account them in the analyses, but estimated them as censored. The number of SA due to O00-O99 was 256, inpatient care due to O00-O99 was 3109 and specialized outpatient care due to O00-O99 was 467.

In the final sample, mean age at baseline in 2005 was 48.6 years (range 16–80, SD 18.2) and 48% were women. Since baseline, 7464 first incident SA spells took place (11% of the final sample) and in the final sample, 42% had inpatient care and 83% specialized outpatient care. The mean follow-up time was 3.2 years (range 0–9.0, SD 3.1 years).

The main diagnoses for inpatient care were I00-I99 (n = 4394), R00-R99 (n = 3042), and S00-T98 (n = 2954), and for outpatient care ICD-10 Z00-Z99 (n = 6980), S00-T98 (n = 5140), and M00-M99 (n = 4543) (Supplemental Table 2). Comparison of inpatient-, outpatient- and no care groups indicated that those with care were older, more often less educated and likely married without children than those without patient care (Table 1).

Figure 2 presents the Kaplan-Meyer survival curves for inpatient care across the SA diagnoses categories and Fig. 3 for specialized outpatient care showing similar patterns for the risk of patient care across the main diagnosis categories of SA over the years of follow-up.

The main SA diagnosis groups (F00-F99, J00-J99, M00-M99, any other diagnoses and missing diagnoses were all strongly associated with increased risk of future care (Table 2) in comparison to no SA. Instead, in comparison to those with missing diagnosis for SA, no SA, SA due to M00-M99, and SA due to any other diagnosis were associated with increased risk of both inpatient and specialized outpatient care (whereas SA due to J00-J99 only with inpatient care). Controlling for confounders (family situation, living area, and education) attenuated the associations in magnitude with retained direction.

Assessment of results of discordant twin pairs (Table 3) shows weak or no effect of familial confounding since the HR were in the same direction and only slightly attenuated among the DZ pairs. Table 4 summarizes the incidence rates and time at risk for the SA main diagnosis groups showing a longer survival time for specialized outpatient care than inpatient care.

This study with 69,552 twin individuals was designed to investigate if SA due to various diagnosis groups differ in the associations with for subsequent inpatient- and specialized outpatient care while controlling for the effect of familial confounding (genetics and early, shared environment). Our results indicate that the main SA diagnosis (F00-F99, J00-J99, M00-M99, any other diagnoses and missing diagnoses) were all associated with increased risk of future hospital care and the association was slightly affected by the confounding factors including familial ones but more among inpatient than in outpatient specialized care. Although SA is common prescription in healthcare [8], the studies of the consequences of SA have been rare. Therefore our study adds to the literature about the link between SA and a number of negative health-related consequences (disease whether or not the same as being granted SA for, or another) [7, 15]. Furthermore, we addressed the consequences of SA at specific, i.e. diagnosis group level to understand the process of SA [15].

Another part of our results focused on testing the group missing SA diagnosis as a reference instead of no SA group. Those results indicated a clear protective role of not having SA for inpatient – or specialized outpatient care. Furthermore, SA due to M00-M99 and SA due to any other diagnosis were associated with both inpatient- and specialized outpatient care. However, our results for other SA diagnosis groups maintained the direction towards higher risk, being indicative that the effect is universal across the diagnosis groups. Taken together, these results do not fully confirm our hypothesis that the associations between SA and patient care would differ by diagnoses or by patient care type i.e. inpatient vs. specialized outpatient care. Instead our results support the theoretical assumption that consequences of SA should be investigated at specific, i.e. diagnosis group level to understand the process of SA [15].

Yet we cannot completely reject our hypothesis. We detected higher incidence rates and survival times for specialized outpatient care than for inpatient care. This might indicate a trend towards severity of disease assumption, i.e. that patient care type plays a role. However, since patient care in terms of inpatient- or specialized outpatient care is often needed for various symptoms and diseases [20, 21], further research is needed to clarify the severity of diseases assumption with more detailed investigation of diagnoses within main ICD-rubrics and also for this association. One may speculate that organization of health care may play a role among those with SA, alternatively self-medication or other self-help may influence the decision of seeking care after SA (towards less contact to health care), stigma related to e.g. mental health symptoms (also towards less contact to health care) or there may be some other influential factors [40‐43]. These factors could not be controlled in this study, but if existing, they might have diluted the results.

The utilization of a twin sample enabled us to control for familial confounding. This adds to epidemiological knowledge based on the individuals in which no genetics or shared (usually early childhood) environmental factors can be controlled for [29, 30]. Our results based on the comparison of the whole cohort to discordant twin pairs showed the point estimates to retain the magnitude and direction, suggesting there is weak or no effect of familial confounding in the associations between SA due to various diagnosis groups and patient care. Hence this independency from familial confounding suggest direct link between SA and patient care although we expected to see familial confounding based on the genetic influence on many chronic diseases that leads to SA [25‐28], but also on SA itself [9, 22]. For (occupational) health care including any preventive actions for SA and/or hospitalizations in terms of inpatient- or specialized outpatient care means both tailored interventions specified by the underlying condition or diagnosis as well as monitoring SA. This arises from the fact that our results point toward an assumption that SA might be an independent and early indicator (i.e. due to weak or no effects of confounders including familial ones) for the need of care. Besides being a common prescription in health care and a part of treatment processes, SA could be utilized as a preventive means for need of care, i.e. being indicative for workplace or individual level actions. These actions could be part of communication based on incidence of SA between health care providers, workplaces and communities for relevant care or prevention.

This longitudinal, population-based register study of Swedish twins has several strengths. The register data is comprehensive, of high quality and without bias or loss due to drop out. We had the opportunity to explore all the main diagnosis groups of ICD-10 both for SA and hospitalizations adding precision. Furthermore, we also tested both no SA and those with SA but missing diagnosis as reference groups to assess the importance of diagnosis which has rarely been done before. However, this kind of register study has also some pitfalls. One limitation was the fact that we only had diagnosis for SA since 2005 limiting the length of the follow-up up to maximal 9 years. Hence one may assume that even longer follow-up might add knowledge on the associations. Another limitation is based on the diagnosis groups. One may assume a tendency of e.g. severe psychiatric disorders to be treated in inpatient care whereas common mental disorders might be addressed in primary health care. This might bias our results due to the underlying disease leading the treatment decisions and potentially being linked with our severity of diseases assumption. Hence even more comprehensive studies with access to high quality register data of primary care and health care level (i.e. primary, occupational, inpatient or outpatient care) of issuing SA for full coverage of all types of care would be needed in the future. The existing register data allowed us to control for several factors in the analyses (i.e. age, sex, education, family status and living area besides genetics and family background). However, we could not control SA before baseline in 2005 which may have overestimated the association since earlier SA are known to predict later SA [44]. Since we utilized education as a proxy for socioeconomic status including employment, we did not account the employment status which could be controlled in further studies. In addition, due to the process in acquiring SA and then seeking patient care, there might be other influential factors related to living situation, working life, personality or else that should be collected via surveys or other means and therefore not available for us. Consequently, we may have overestimated the associations and further studies should confirm if this was the case. The Swedish sample with register data from a welfare country with SA benefits and patient care for all citizens [36] can limit the generalizability of our results. We though assume that these findings can be useful within Nordic countries with similar welfare status and indicative also for other countries for adding to the knowledge for tailoring interventions for underlying conditions or diagnosis.

Sickness absence predicts both inpatient– and specialized outpatient care and these associations were universal across diagnosis groups. On the other hand, the survival time and incidence rates of inpatient care were lower than for specialized outpatient care pointing towards the assumption of severity of diseases although this finding was universal across diagnosis groups of sickness absence.