The data were a part of the National Health Screening Service in Norway and were collected in the Nordland County from August 1988 to March 1989. Individual-level information was obtained from a database of national insurance, created by Statistics Norway and the Norway National Insurance Service. Follow-up time was from January 1, 1992 to December 31, 2007. The study was approved by the Regional Committee for Medical Research Ethics (2009/205-4).
Nordland County is one of 19 counties and is situated in the northern part of Norway. In 1990, Nordland County had 45 municipalities and 239,532 inhabitants. In Statistics Norway's categorization, expressed in terms of the relative distribution of industries in relation to the working population residing in the municipalities in 1990, Nordland County had municipalities where the main industries were fishing, agriculture, manufacturing and services. The diverse types of industries in the municipalities were likely affected differently by business fluctuations during the follow-up period.
Disability pension was established to ensure sufficient income for people whose earning ability is permanently impaired by at least 50% due to illness or injury. Although each insurance office can exercise some discretion in their decisions, and thus be more lenient to people who have obvious problems finding new jobs, the law requires a medical diagnosis. In this study, the dependent variable was the first day of work disability, defined as the time when a person's earning ability was permanently reduced. In most cases, this date represents the first day of long-term sickness benefits for persons who were later granted a disability pension. Data on new incidents of disability pensions were available from January 1, 1992, and covered all cases of disability pensions in Norway. No cases were missed in this period as firm and private disability insurance is always supplementary to the national pension.
The impact of unemployment was hypothesized to influence the subsequent risk of disability pension with some induction time. Hence, assessing work disability after unemployment was done as a time-varying covariate with a one-year time lag, meaning the risk of work disability is measured one year after becoming unemployed. Participants were classified as unemployed the year they started an unemployment period. With sensitivity analyses, we also tested models without a time lag of unemployment and with a two-year time lag. Data were obtained from the national insurance register.
Baseline information on different aspects of health was used to adjust for health impairment prior to unemployment. A summated index of the number of chronic illnesses included the following conditions: myocardial infarction, angina pectoris, stroke/cerebral infarction, diabetes, high blood pressure, chronic bronchitis, arthritis, Bechterew's disease, cancer, epilepsy, migraine and gastro-intestinal problems. Self-rated health status was assessed by the question, "what is your health condition like?" The question had four answer categories: "Very good", "Good", "Fair" and "Poor". Depression was assessed by the question, "have you been sad or depressed the last 14 days?" The four answer categories ranged from "almost all the time" to "never or rarely". Headache and pains in the neck and shoulders were measured with a four-point scale, ranging from "never/rarely" to "daily". Alcohol use was assessed with a four-point scale, ranging from "non-drinker" to "daily drinker" Smoking was assessed with a three-point scale with the responses of "non-smoker," "former smoker" and "smoker".
The age of the participants was between 40-42 years at baseline. Education level was used as a measure of socioeconomic status and included the three categories, "primary school", "high school" and "college/university".
The association between unemployment and disability pension was estimated with discrete time multilevel logistic regressions with individuals nested by municipality of residence. In a discrete time logistic regression analysis, time is treated as intervals, and the risk of disability pension (event) is measured within each interval, given that the event has not occurred before [21
]. We used one-year intervals that corresponded with calendar years. The risk of receiving disability pension is closely related to age [22
], and therefore, we used age during follow-up period and age-squared to assess the combination effect of age and follow-up period.
In order to explore the impact of individual municipalities, we estimated a conditional Intra- class correlation coefficient
]. For the present study, the ICC provides an estimate of the relative importance of the municipality location on an individual's propensity to receive disability pension.
The association between unemployment and subsequent disability was performed in three models. Model 1 was adjusted only for age (i.e., age and period) and sex. In Model 2, we also included baseline health status, health behaviour (as measured by alcohol and smoking behaviour). In Model 3, education was added to Model 2. The precision of the estimates was represented by 95% confidence intervals (CI). The analyses were limited to the participants with complete information in all study variables (5,834). All analyses were conducted using STATA 11 software (StataCorp LP, Texas, USA).
Effect measure modification analysis
We tested statistical interactions among the variables to investigate the effects of age in follow-up, sex and level of education on the unemployment-disability pension odds ratio.