Intervention studies are needed in occupational health to test whether work improvements or changes in working conditions are effective. Nevertheless, interventions, especially randomized interventions, are not always feasible. In the last decade, propensity score analysis has been widely used in clinical settings for evaluating the role of “treatments” for which randomized controlled trials were unlikely to be conducted, such as changes in socioeconomic environment in public health. The propensity score method consists in an estimation of the average treatment effect by making treated and untreated group comparable for confounding factors.1 The propensity score is defined as the inverse of a subject probability of receiving a treatment, generally unknown in observational studies. The propensity score is estimated by using a multinomial logistic model on confounding factors between treatment and outcome. Then, the treatment is considered with the propensity score (matched, adjusted, or weighted) to estimate the effect of treatment if treated and untreated subjects would have had the same distribution of covariates.2 Propensity score analyses are feasible for essentially any treatment or intervention with a finite number of possibilities. We aimed to describe the frequency of articles using propensity score cited in occupational health literature and its use. Using the key words “occupational” and “propensity score,” studies published in the last 20 years were selected from PubMed database. Frequency of citation was compared with two other requests: first, “clinical” and “propensity score”; and second, “public health,” not “occupational,” and “propensity score.” Embase and Web of Science were also used to find extra articles with similar key words. Relevant information on the use of propensity score was extracted from the selected articles. Twenty-four articles corresponding to the selected criteria were found in PubMed, of 234,523 references in occupational health in the 20 years (search performed on September 20, 2012), which corresponded to 0.01% of all citations in the period in that area. More references were found: 1210 references of 2,804,712 (0.06%) in clinical domain and 2731 of 4,982,627 (0.07%) in public health domain. Both frequencies were significantly different from that observed in occupational health (chi-squared test, P < 0.001 for both tests). Among articles in occupational health, 12 were relevant, published since 2005, mostly from US teams (n = 10; Table 1).3–14 The analyses included many subjects (median, 7314), coming from preexisting databases. The studies dealt with treatment evaluation (n = 2), economic evaluation (n = 2), and risk factors analysis (n = 5). Few studies used propensity score analysis for interventions; two used this method for evaluating interventions (safety or coaching programs), and only one of them really considered work adaptations or rehabilitations.TABLE 1: Relevant Citations Using Propensity Score Analyses in Occupational HealthThe review presented here has limitations. Because the search was restricted to “propensity score” key term, some relevant references may have been missed. Nevertheless, taking into account that the proportion of missing reference in occupational health should probably be similar to that in other field, it is unlikely to be an important issue. Propensity score analyses are particularly useful when randomized controlled trials are unlikely to be conducted. Nevertheless, investigators need to dispose of enough statistical power and relevant variables to assess the propensity of “treatment.” In particular, the choice of variables in propensity score is crucial to obtain an efficient estimation of average treatment effect,15 and propensity analyses are particularly adapted in large samples. In conclusion, more widespread use of this methodology in large workers data sets might give information on efficiency of work adaptation where intervention studies are not suitable or feasible. Alexis Descatha, MD Univ Versailles St-Quentin, Versailles, France; Inserm, Centre for Research in Epidemiology and Population Health, “Population-Based Epidemiological Cohorts” Research Platform, Villejuif, France; and AP-HP, Occupational Health Unit/EMS (Samu92), University Hospital of West Suburb of Paris, Garches, France. Annette Leclerc, PhD Univ Versailles St-Quentin, Versailles, France; and Inserm, Centre for research in Epidemiology and Population Health, “Population-Based Epidemiological Cohorts” Research Platform, Villejuif, France. Eleonore Herquelot, MSc Univ Versailles St-Quentin, Versailles, France; and Inserm, Centre for research in Epidemiology and Population Health, “Population-Based Epidemiological Cohorts” Research Platform, Villejuif, France.