The data file „mendepublicuse180220.dat“ includes the data for the replication of the study results in a fixed ASCI format. The file “mendepublicuse180220.sts” includes the data definition.  
Since our study participants are psychiatric patients treated for depression we cannot share individual data on age, education, income, employment status, partnership and children for privacy reasons. However, we provide propensity scores representing the individual probabilities of each participant to be assigned to the three latent classes conditional to their individual combinations of these variables. This enables the user of the data set to replicate our study results without access to individual socieconomic information. In our view this procedure represents an adequate balance between scientific transparency and patient privacy rights.
Propensity scores [1] have been estimated on the basis of a multinomial logit regression model using latent class assignment as dependent variable with assignment to latent class three as reference categorie.  The independent varaibles included in the model were: Age (in years), deducation (0= below Abitur; 1 = Abitur and higher), monthly net household income (0 = below 3,000 €; 1 = 3,000 € and higher); unemployed (0 = no; 1 = yes), blue-collar worker (0 = no; 1 = yes); living with a partner (0 = no; 1 = yes); living with children (0 = no; 1 = yes), recuitment setting (1 = psychiatric hospital; 2 = psychosomatic hospital; 3 = family doctor; 4 = press invitation). The multinomial logit regression model has been computed with the procedure “mlogit” in STATA 15. The results of the multinomial logit regression are presented in Table 3 of the manuscript.
The propensitiy scores pcl1 – pcl3 in the data file represent the individual probabilities of the study partcipants to become assigned to the latent classes one, two or three conditional to their combination of the independent variables described above. The propensity scores can be used to estimate adjusted means of the PHQ-9, GAD-7, the PHQ-15, DSS personal, DSS perceived and the DUI broken down by the latent classes presented in figure 2 and figure 3 of the manuscript by including two of the tree variables in a linear regression mode [1].
 
References
1.	Austin PC (2011) A Tutorial and Case Study in Propensity Score Analysis: An Application to Estimating the Effect of In-Hospital Smoking Cessation Counseling on Mortality. Multivariate.Behav.Res. 46(1):119–151