Objectives To examine the longitudinal relationship between incidence of diagnosed chronic disease and work status and hours worked.
Methods A dynamic cohort approach was taken to construct our study sample using the Canadian National Population Health Survey. Participant inclusion criteria included being employed and without a chronic health condition in the survey cycle prior to diagnosis, and participation in consecutive surveys following diagnosis. Each respondent was matched with up to 5 respondents without a diagnosed health condition. The direct and indirect associations between chronic disease and work status and hours worked following diagnosis were examined using probit and linear regression path models. Separate models were developed for arthritis, back problems, diabetes, hypertension and heart disease.
Results We identified 799 observations with a diagnosis of arthritis, 858 with back pain, 178 with diabetes, 569 with hypertension and 163 with heart disease, which met our selection criteria. An examination of total effects at time 1 and time 2 showed that, excluding hypertension, chronic disease diagnosis was related to work loss. The time 2 effect of chronic disease diagnosis on work loss was mediated through time 1 work status. With the exception of heart disease, an incident case of chronic disease was not related to changes in work hours among observations with continuous work participation.
Conclusions Chronic disease can result in work loss following diagnosis. Research is required to understand how modifying occupational conditions may benefit employment immediately after diagnosis.
Statistics from Altmetric.com
What this paper adds
Little research has examined the longitudinal relationship between a chronic disease diagnosis and changes in work status or work hours; existing longitudinal studies have tended to comprise samples with a pre-existing chronic disease and attributing changes in work participation to chronic disease can be difficult.
This study used a dynamic cohort approach to create a sample of participants from a National Population Health Survey who were employed and did not have the chronic disease diagnosis prior to the first analytical cycle. Findings showed that a chronic disease diagnosis was associated with a loss of work status in the first survey cycle following diagnosis and after 2 years, but not with changes in the number of work hours.
Longitudinal path analysis models developed in this study found that work loss in the survey cycle following a chronic disease diagnosis mediated work status after 2 years.
There is a need for early work disability prevention initiatives following chronic disease diagnosis to prevent early labour market exit and encourage sustained involvement at work.
Chronic disease can have a significant impact on involvement in the labour market. In Canada, it is estimated that approximately half of the working-aged adults aged 20–64 years are living with a chronic disease; and four of the five are at risk of developing a chronic condition over their lifetime.1 ,2 The number of people living with a chronic disease are expected to increase dramatically over the next decades, a trend that has been attributed to an ageing population and rise in risk factors.3 ,4 Not surprisingly, studies have also established a clear association between a chronic disease diagnosis and labour market participation restrictions. In particular, findings show that a common chronic condition such as arthritis, diabetes, back problems, hypertension or heart disease is often associated with not working, working fewer hours or reduced productivity, compared with not having each condition.1 ,5–10 At the same time, limited information exists regarding the timing between the onset of a chronic condition and its impact on involvement in the labour market. To fill knowledge gaps, this study used a longitudinal design to examine the association between a chronic disease and work status and work hours at the time period following diagnosis, and after 2-years. Gaining an understanding of temporality will allow researchers to determine whether work loss can be attributed to chronic disease diagnosis. Additionally, findings from our study will provide insight into the best possible time points at which accommodations should be provided to prevent work disability.
Most studies examining the relationship between chronic conditions and labour market participation have been descriptive, using primarily cross-sectional study designs. For instance, in a survey of over 2000 adults aged 25–54 years, 17.5% indicated at least 1 day of work loss and 20.2% reported at least one work cutback day (eg, half day lost) related to a chronic health condition.5 Relatively few studies have used a longitudinal approach to understand work-related changes associated with a chronic disease. Those that do exist have consisted of samples with pre-existing chronic conditions at baseline. One investigation of over 7000 older adults collected over 2 years showed that a self-reported diabetes diagnosis was associated with a higher probability of work loss (4.4% for women and 7.1% men) when compared with their counterparts without the health condition.11 In that same study, diabetes was not significantly associated with a change in work hours, and could suggest that someone with a chronic condition may be more likely to leave the labour market rather than adapt their working conditions.11 A recent 2-year Dutch longitudinal study of close to 11 000 middle-aged and older-aged adults examined how different chronic health conditions influenced work ability and work productivity, when compared with those without a health condition.12 Findings showed that those living with the health conditions examined were more likely to report lower work ability at follow-up.12 Despite the insights provided by existing research, no studies have examined the time point at which a chronic disease is likely to impact work status and hours worked following diagnosis. Examining temporal ordering can allow us to determine whether changes in work participation can be attributed to a chronic disease diagnosis.13 If labour market changes occur immediately after chronic disease onset, then, providing important confounders have been included in an analysis, changes in work participation are more likely to be due to a chronic disease. A longer time period between chronic disease diagnosis and changes in work participation may make attribution more challenging.13
The objective of this study is to examine the longitudinal relationship between the incidence of diagnosed chronic disease and changes in work status and hours worked. Using data from a longitudinal Canadian population health survey, the current study applied a dynamic cohort approach to answer two research questions: does a chronic disease have an impact on work status immediately and two years following diagnosis, and is chronic disease associated with changes in work hours among persons with continuous work participation? Relative to age-matched observations without a chronic condition, we hypothesise that patients receiving a chronic disease diagnosis will be at greater risk of work loss, but will not report working fewer hours.
For this study, we used the Canadian National Population Health Survey (NPHS). NPHS is a longitudinal survey that was conducted every 2 years between 1994 and 2010 (when the survey was concluded). The original baseline cohort of the NPHS totalled 17 276 unique respondents, and the response rate ranged from 92.8% in cycle 2 to 69.7% in cycle 9.14 Our analysis aimed at comparing the change in work status and hours worked for the respondents who were employed and had an incident case of one of the chronic diseases of interest (ie, arthritis, back pain, diabetes, hypertension or heart disease) versus those who were employed but did not develop one of these conditions. Chronic conditions examined were selected based on their high population prevalence, as well as their relationship with age and employment participation established in previous literature.15 We compared the effects of the incident condition on work status and hours worked contemporaneously, and 2 years later. This required the data set to have three consecutive observations: time 0=when respondents were employed and free of the condition; time 1=when the condition was first reported within the 2-year survey cycle following time 0; and time 2=survey cycle 2 years after the cycle in which the condition was first reported.
A dynamic cohort approach was taken to construct our study sample. The analytical data set was set up separately for each condition using three steps. First, we identified respondents who reported the condition of interest at any point between 1996 and 2008. That meant that respondents did not have the health condition in the NPHS first cycle (1994) and had at least one follow-up survey cycle after the first report of their condition. Also, we restricted our sample to respondents who were working in the cycle preceding their first report of the condition, and indicated a chronic disease diagnosis at T1 and T2. Overall, all participants had three consecutive responses to the NPHS, and must have been working and without the condition of interest in the first analytical time period.
In the second step, each of our respondents who met our chronic condition eligibility criteria were matched with up to five respondents not indicating an incident case of the condition of interest. Respondents with the condition and those without were matched based on survey cycle, age group (in 10-year increments) and gender. Similar to respondents with the condition, all respondents without the condition of interest had to be working in the first period in which data were collected and have consecutive responses to the NPHS. Once a respondent without a condition was selected as a match for a respondent with an incident case, they were excluded from being a comparison for other respondents with this same health condition in subsequent cycles. Using the dynamic cohort approach, we were able to increase the number of respondents with and without conditions of interest, while also controlling for labour market participation restrictions attributed to external factors occurring over time (eg, economic conditions). Based on the process of selecting respondents, each model may consist of a different subsample of the overall cohort and overlap.
Third, the main outcome used in this study was the working status question of the NPHS, which compared those who were ‘not working because of health reasons’ to those who were ‘working’. We included participants who reported working full time and part time. We excluded respondents, with and without incident cases of a health condition, who were in other work status categories including ‘retired’ and ‘not working because of family reasons’. We also removed respondents who had missing information on our covariates of interest, or did not have information on the status of the condition of interest in the second or third analytical survey cycle. A complete description of attrition through each of these steps for respondents with incident cases of each condition is reported in table 1.
It is important to acknowledge that we have not used a case/control methodology in our analysis (eg, conditional logistic regression). Accordingly, we do not refer to respondents with the conditions as ‘cases’ and respondents without the condition of interest as ‘controls’. Instead, we have undertaken path modelling to examine whether the impacts of a chronic health condition on work status or work hours occur in the first survey cycle following diagnosis and after 2 years. In addition, path modelling was used to examine whether work status changes or work hour changes in the time 1 survey cycle were associated with work status or work hour changes in the time 2 survey cycle. Undertaking conditional regression approaches is not possible with this analytical approach.
Outcome: work status and work hours/week
We were interested in two labour market outcomes. The first was working status (‘working’ vs ‘not working due to a health reasons’) that was ascertained in the NPHS by asking ‘Last week, did you work at a job or business?’.12 The second outcome was number of work hours/week, assessed using the questions ‘About how many hours a week do you usually work in your job or business?’. An examination of work hours was restricted to a subset of cohort members who reported working in the survey at time 1 and time 2. For this model, we examined work hours in the cycle subsequent to the report of a health condition, adjusting for work hours in the cycle before the disease diagnosis. Owing to the range of response, work hours were capped at 80 hours/week. An examination of work hours indicated that the variable was normally distributed.
Independent variables: incident chronic conditions
Incident chronic conditions were the primary independent variables investigated in the study. We examined each of the following commonly reported conditions in separate analyses and data sets: arthritis, back problems, diabetes, hypertension and heart disease. The incidence of each condition was self-reported by the respondent. For each survey cycle, chronic conditions were defined as ‘long-term conditions which are expected to last or have already lasted 6 months or more and that have been diagnosed by a health professional’. Respondents were asked about the presence of each of the five conditions explicitly.
Models were adjusted for several covariates that were measured in the first cycle. Covariates were chosen based on their association with work status and work hours and chronic disease diagnosis in previous research (1), including gender, age, marital status, body mass index (based on self-reported height and weight), living in urban/rural setting, presence of comorbidity (ie, any 1 of the 5 conditions examined, aside from the condition of interest within the specific chronic disease model) and depression status (self-reported Composite International Diagnostic Interview16). Work-related factors included were also examined including working hours and minimal occupational strength requirements which were determined by mapping a predefined categorisation to the occupation reported by a particular case (limited=handling loads up to 5 kg; light=handling loads of 5 kg but <10 kg; medium=handling loads between 10 and 20 kg; heavy=handling loads >20 kg).17
Analyses were undertaken using path models, which enabled an examination of work experiences ∼2 years following the survey cycle in which the condition was diagnosed (time 2), both directly (ie, independent of the immediate impacts of the condition on labour market participation) and indirectly through the impacts of the chronic condition on work status and hours worked in the first survey cycle following diagnosis (time 1). Path models also provide fit indices that allow us to determine how well our model fits the data. An overview of the path model used for the analysis is presented in figure 1. The primary pathways of interest in this analysis are those examining the time 1 and time 2 impacts of an incident case of a given chronic condition on work participation. For each path model, the direct time 1 impact of the condition is estimated by path A, and the direct time 2 impact is estimated by path C. The indirect time 2 impact is the combination of paths A and B (ie, the impacts on time 2 work participation that are mediated through the time 1 effect of the condition on work status or hours worked).
We ran separate path models for work status and work hours. When examining the probability of work loss due to health reasons at time 1 and time 2, we undertook probit regression analyses, given that the outcome was dichotomous. Probit regressions are comparable in their interpretation to more traditional logit estimates, and provide similar conclusions. It is suggested that probit estimates can be multiplied by 1.6 to approximate logit estimates.18 ,19 Next, we performed linear regression analyses to examine the association between a chronic health condition and work hours at time 1 and time 2. The sample for the analyses examining work hours was restricted to only those participants still in the labour market. Sampling weights, which account for initial probability of selection and unit-level non-response to the initial NPHS survey, were included in all analyses as recommended by Statistics Canada.14 Data preparation was performed using SAS V.9.3 (SAS Institute. SAS V.9.3. Version 9.3 ed. Cary, North Carolina, USA: SAS Institute; 2015). Path models were conducted using Mplus.20 The study protocol was reviewed by the University of Toronto Health Sciences Ethics Review Board.
Initially, 8434 observations indicating a chronic disease diagnosis were examined for study inclusion. As exhibited in table 1, 2547 observations were working in the cycle prior to the onset of their condition, and had a self-reported diagnosis of arthritis (n=779), back pain (n=858), diabetes (N=178), hypertension (n=569) or heart disease (n=163). Each observation was matched with up to five observations without a chronic condition based on age, gender and survey cycle. Demographic characteristics, working conditions and health factors of cases with and without a chronic disease diagnosis are summarised in table 2. The presence of a comorbidity ranged from 23% to 32% across the different chronic disease models. Information on the survey cycle in which a person was diagnosed with a chronic condition and Canadian Province in which a participant resided is presented in online supplementary appendix table S1.
supplementary appendix table
Survey wave in which a participant was diagnosed with a chronic condition and Province resided. Weighted N includes all participants with and without diagnosis in each model.
Table 3 presents the adjusted probit regression estimates for the path models and total effects for time 1 (path A) and time 2 (path C+(path A×path B)). Also presented in table 3 is the time 2 direct (path C) and indirect effects (path A×path B) of an incident case of arthritis, back pain, diabetes, hypertension or heart disease on work status. Analyses are presented separately for each health condition, since models involve a different data set of observations. After adjusting for covariates, the models showed that with the exception of hypertension, a chronic disease diagnosis had a significant total effect on work loss both at time 1 and time 2. Findings from the path model also showed that the time 2 effect of a diagnosis of chronic disease on work status was indirect, mediated through time 1 work status. Those who were working within the first survey cycle following diagnosis of a health condition were more likely to be employed in the subsequent 2-year survey cycle. Diabetes and heart disease exhibited the largest total effects at time 1 and time 2. Given the dynamic cohort approach, each model consists of a different subsample of the overall cohort, and each of these subsamples may overlap to varying extents. Accordingly, estimates should be compared with caution.
Table 4 presents the adjusted linear regression estimates for the path models examining total effects, and direct and indirect effects of a chronic condition on work hour changes at time 1 and time 2. In these models, work hours at time 1 and time 2 are the outcome, adjusting for work hours at time 0. Negative coefficients in each model should be interpreted as working fewer hours than anticipated given the work hours at time 0; positive coefficients should be interpreted as working more hours than anticipated at follow-up given the work hours at time 0.21 Models were stratified based on the health condition. After adjusting for covariates, only heart disease had a significant effect on work hours at time 1 and time 2 among persons with continuous employment. Similar to working status, results showed that the time 2 effect of a diagnosis of chronic disease on work hours was indirect, mediated through time 1 work hours. An incident case of any of the remaining chronic health conditions did not have a total effect on changes in work hours at either time 1 or time 2.
Path models presented in tables 3 and 4 exhibited adequate fit when compared with a model assuming the variables are uncorrelated (Tucker-Lewis index >0.95, comparative fit index >0.90), and based on residual fit-based indices (root mean square error of approximation (RMSEA) <0.05).
Chronic disease diagnosis has a significant impact on involvement in employment. Using a dynamic cohort approach, we conducted a longitudinal analysis of the NPHS to examine the association between the incidence of one of five health conditions including arthritis, back problems, diabetes, hypertension or heart disease and changes in work status, and work hours in a representative sample of Canadians. All conditions diagnosed in the two survey cycles, with the exception of hypertension, were associated with work cessation. Using path models, the findings also showed that the time 2 effect of a chronic disease diagnosis on work status was mediated by work cessation in the first survey cycle in which the disease was diagnosed. Among those who continued working, with the exception of heart diseases, a chronic disease diagnosis was not associated with changes in the number of work hours at time 1 or time 2. When taken together, the results may reflect a potential all-or-nothing effect, where an incident chronic condition in our cohort was associated with leaving the labour market altogether and not related to changes in work hours. The findings may indicate the potential for early intervention following disease diagnosis to identify appropriate accommodations as a means to prevent labour market exit and sustain involvement in employment.
Corroborating a body of research, our study shows a significant association between chronic disease diagnosis and work loss.5 ,11 ,12 In the light of an ageing workforce and rise in the number of chronic disease-related risk factors, the findings bring further attention to the importance of understanding the workplace accommodations and supports required to minimise potential work limitations and encourage sustained employment. Notably, our study examined a path model, to determine the association between chronic disease and work loss at two time points. A chronic disease diagnosis resulted in work loss both at time 1, and the subsequent survey cycle, approximately 2 years later (time 2). On the basis of our temporal examination, we may be able to attribute the change in time 1 work status to the chronic disease diagnosis.13 The findings also show that the time 2 effects of acquiring a chronic disease on work status were mediated by work loss in time 1. The early phase following chronic disease diagnosis is an important period that requires timely strategies to prevent work disability. For instance, there may be a need for the first point of contact in an individual's circle of care (eg, primary care or occupational health physicians) to help with chronic disease management, and the minimisation of work-related barriers to prevent labour market exit.22 To tailor work disability prevention initiatives to different phases of an individual's health condition, our study also suggests the requirement for research to understand which personal, health or organisational factors may be most related to early and prolonged loss of employment following a chronic disease diagnosis.23
With the exception of heart disease, chronic diseases examined in this study were not associated with changes in the number of hours a person worked early and 2-years following diagnosis. Our findings align with previous studies, which also show that chronic disease diagnosis can be associated with work loss, but not with adapting work hours.11 The results could reflect an economic need associated with the diagnosis of a chronic disease. Individuals may not be changing the hours they work to maintain access to income, and benefits required to manage their health condition.3 ,24 On the other hand, a diagnosis of heart disease may be more likely to be followed by recommendations to limit the volume or strenuousness of work.25 The findings might also indicate gaps in the availability or access to accommodations required to sustain employment with a chronic disease. A recent investigation of work experiences of people living with arthritis showed that despite their perceived need and availability, respondents were not always accessing the job accommodations or modifications that they required most; not accessing accommodations was associated with disruptions to work.26 Additional research is required to determine whether individuals with different chronic diseases require modifications to their working conditions to sustain employment. Among those needing modification, studies should also be conducted to determine the ways in which the workplace can be accommodated to address potential barriers. These insights are vital to the development of workplace policies and programmes that can prevent work loss and sustain employment.
Our study had several methodological strengths. Notably, we used a dynamic cohort approach to develop a sample of participants with and without a chronic health condition, and analysed change in work status and work hours longitudinally while controlling for a range of covariates. Through our study design, we were able to analyse a sample of labour market participants who were all free of a particular health condition, and examine the impact of the onset of these conditions on work status and work hours over two subsequent survey cycles. Limitations should also be acknowledged. The NPHS uses a self-report of medical diagnosis of a chronic condition, which may be a source of measurement error when compared with administrative records of objective chronic disease diagnosis.27 This source of variation may be especially salient among those living with a health condition that has not been clinically diagnosed or those who may incorrectly indicate having a health condition as a way to justify changes in work status.27 In the NPHS, employment information was collected over multiple survey cycles administered every 2 years. Within each survey cycle, temporal ordering is not ascertained, and it is uncertain whether changes in work status extend beyond the analysis time period. Work outcomes including absenteeism and presenteeism may be accessed in future research to enhance our understanding of the changes in work participation associated with a chronic disease diagnosis. While important covariates were considered in our analyses, we also could have controlled for additional demographic and work-related factors. Subsequent research could also aim at matching respondents with and without a chronic disease diagnosis based on a broader range of factors. In each model, we examine the association between an incident case of a specific chronic disease and changes in work status or work hours. In these models, the presence of a comorbidity was treated as a covariate. Using a similar dynamic approach to constructing a cohort, future research could examine how the incidence of comorbidity could influence work status or hours worked over multiple time points.
In the light of an increasing number of people living with a chronic disease, there is a need for research to better understand how a diagnosis of a chronic condition may influence work participation. Using a longitudinal design, our study showed temporality in the effect of a chronic disease diagnosis on work participation. Early work loss may be attributed to a chronic disease and persist for at least 2 years following diagnosis. Incidence of chronic disease was not related to changes in the number of hours a person may work. While a chronic disease diagnosis may increase the need for resources provided by full-time employment, individuals may not be adapting their occupational circumstances to sustain employment. There is a need to better understand how both early intervention and changes in working conditions for individuals with chronic disease can prevent changes in work status, and ensure successfulness at work.
Contributors AJ, CC, CM, SI, DB, AB and PS conceived the study idea, significantly contributed to data analysis procedures, helped with the interpretation of findings and assisted with manuscript development. AJ led the manuscript development. CC and PS co-led the analysis.
Funding This project was supported with funds from an operating grant from the Canadian Institutes for Health Research (grant number 111273).
Competing interests PS was supported by a Discovery Early Career Researcher Award from the Australian Research Council, and is currently supported by a Canadian Institutes of Health Research, Research Chair in Gender, Work and Health.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.