Risk reclassification analysis investigating the added value of fatigue to sickness absence predictions

Risk reclassification analysis investigating the added value of fatigue to sickness absence predictions Corn A. M. Roelen 0 1 Ute Bltmann 0 1 Johan W. Groothoff 0 1 Jos W. R. Twisk 0 1 Martijn W. Heymans 0 1 0 C. A. M. Roelen U. Bultmann J. W. Groothoff Department of Health Sciences, University Medical Center Groningen, University of Groningen , Groningen , The Netherlands 1 C. A. M. Roelen J. W. R. Twisk M. W. Heymans Department of Epidemiology and Biostatistics, VU University Medical Center, VU University , Amsterdam , The Netherlands 2 ) ArboNed , PO Box 158, 8000 AD Zwolle , The Netherlands Background Prognostic models including age, self-rated health and prior sickness absence (SA) have been found to predict high (30) SA days and high (3) SA episodes during 1-year follow-up. More predictors of high SA are needed to improve these SA prognostic models. The purpose of this study was to investigate fatigue as new predictor in SA prognostic models by using risk reclassification methods and measures. Methods This was a prospective cohort study with 1-year follow-up of 1,137 office workers. Fatigue was measured at baseline with the 20-item checklist individual strength and added to the existing SA prognostic models. SA days and episodes during 1-year follow-up were retrieved from an occupational health service register. The added value of fatigue was investigated with Net Reclassification Index (NRI) and integrated discrimination improvement (IDI) measures. Results In total, 579 (51 %) office workers had complete data for analysis. Fatigue was prospectively associated with both high SA days and episodes. The NRI revealed that adding fatigue to the SA days model correctly reclassified workers with high SA days, but incorrectly reclassified workers without high SA days. The IDI indicated no improvement in risk discrimination by the SA days model. Both NRI and IDI showed that the prognostic model predicting high SA episodes did not improve when fatigue was added as predictor variable. Conclusion In the present study, fatigue increased falsepositive rates which may reduce the cost-effectiveness of interventions for preventing SA. Absenteeism; Prediction models; Reclassification table; Risk distribution; Risk stratification; Sick leave - Sickness absence (SA) is an increasing problem in developed economies. The Organization for Economic Cooperation and Development reported that countries spend on average 2 % of their gross domestic product (GDP) and 10 % of their social expenditures on SA and disability benefits (OECD 2011). When off work due to sickness, the probability of resuming work decreases with increasing SA duration (Labriola 2008; Lund et al. 2008). Eventually, long-term SA leads to disability pension which excludes workers from the workplace and marginalizes them from the labor market. Therefore, it is important to identify high-risk workers before they report sick, so that they can be invited for interventions aimed at preventing SA (Taimela et al. 2008a; Kant et al. 2008). Recently, two prognostic models for identifying workers at risk of high SA were developed in a sample of Dutch hospital workers (Roelen et al. 2013a) and validated in Dutch office workers (Roelen et al. 2013b) and Danish eldercare workers (Roelen et al. 2014a). The prognostic model predicting high SA days (i.e., 30 cumulated days during 1-year follow-up) showed fair performance in hospital workers, but poor performance at external validation. The prognostic model predicting high SA episodes (i.e., 3 episodes during 1-year follow-up) showed good performance in the development setting and maintained fair performance at external validation. It was concluded that more predictors of SA are needed to improve the SA prognostic models, particularly the model predicting high SA days. Fatigue is a common symptom of ill-health, ranking third in prevalence after back pain and muscular aches in working populations (Parent-Thirion et al. 2012). Wikman et al. (2005) reported that fatigue was the most prevalent indicator of morbidity in the Swedish workforce. Several studies have already shown that fatigue is associated with future SA (Janssen et al. 2003; Bltmann et al. 2005, 2013; kerstedt et al. 2007; Roelen et al. 2014b). However, we need more research to investigate whether or not fatigue should be added as predictor to the SA prognostic models. Traditionally, the added value of a new predictor to existing prognostic models is investigated by changes in the area under the receiver operating characteristic curve or c-statistic (Steyerberg et al. 2010). However, only very strong predictors can increase the performance of wellpredicting prognostic models (Pepe et al. 2004; Janes et al. 2008; Cook 2007, 2008). The Net Reclassification Index (NRI) has been introduced as novel measure to assess the added value of predictor variables (Pencina et al. 2008; Steyerberg et al. 2010; Cook and Paynter 2011; Sundstrm et al. 2011). The NRI summarizes reclassifications of subjects when the new predictor is added. Subjects who develop the outcome are correctly reclassified when they move up into a higher risk category, and subjects who do not develop the outcome are correctly reclassified when they move down into a lower risk category. A disadvantage is that the NRI heavily depends on the thresholds of risk stratifications (Sundstrm et al. 2011). Pencina et al. (2011) defined an NRI without using risk categories, but this category-free NRI variant has been criticized for its high rates of false-positive conclusions (Pepe et al. 2014). The integrated discrimination improvement (IDI) is an alternative category-free measure to quantify risk discrimination improvement (Pencina et al. 2008; Steyerberg et al. 2010; Sundstrm et al. 2011). Empirical evaluations of the literature showed that reclassification methods and measures are often applied inappropriately, for example, to assess the performance of new prognostic models that differ from the previously established models (Tzoulaki et al. 2011; Bouwmeester et al. 2012). In addition, the reclassification measures NRI and IDI are frequently misinterpreted (Kerr et al. 2014; Leening et al. 2014a). The objective of the present study was to introduce risk reclassification and related NRI and IDI measures in an occupational health context by investigating the added value of fatigue to the existing SA prognostic models. Reclassification analysis can be a key method for guiding decisions in occupational health care, because it presents the distribution of risks in the population and classifies subjects into relevant risk categories. To illustrate risk reclassification, we used the health check data from a previously described prospective cohort study of 1,137 office workers (Roelen et al. 2013b, 2014b). Predictor variables were measured by health check questionnaires administered in November 2006. SA in 2007 was retrieved from an occupational health service (OHS) regi (...truncated)