Trajectories of somatic drug utilization patterns in Major Depressive Disorder: A Study protocol for a Danish nationwide register study using Latent Class Analysis

Study design and setting

The study will be a register-based drug utilization study assessing somatic pharmacological treatment profiles and their trajectories, or transitions over time, in patients with a first-time MDD diagnosis. The MDD patients are diagnosed at time 0 (index date) between 2011 and 2015 (the study inclusion period) and followed over the patient observation period defined as three years prior to the index MDD diagnosis until three years after the diagnosis (Figure 1A). Using Latent Class Analysis (LCA) the somatic drug profiles for the MDD patients will be determined at four seven-month intervals defined by the following time points: three years before the index date, index date, seven months after the index date, and three years after the index date (Figure 1B). The trajectories, i.e. drug profile changes throughout the four periods, will then be determined and characterized.

Figure 1. A) Illustration of the patient observation period depending on the index date between 1^st of January 2011 and 31^st of December 2015 (study inclusion period). B) Somatic drug profile is determined for the patients during the 6-year total patient observation period.

Study sample

Inclusion criteria

To be eligible for the study, individuals must have met three inclusion criteria: 1) Patients have received an MDD diagnosis at a psychiatric unit between 1^st of January 2011 till 31^st of December 2015. 2) The MDD diagnosis must be a first-time (index) diagnosis. The diagnosis is defined as first-time if the patients do not have a history of MDD 15 years prior to the (index) diagnosis. 3) The MDD diagnosis must be the primary/action diagnosis (“A-diagnosis”).

The information regarding the MDD diagnosis, date of diagnosis, and diagnosis type is available through the Danish National Patient Register (DNPR).¹¹ The International Classification of Diseases, Tenth Revision (ICD-10) will be used to identify the MDD diagnosis including either a single episode of depression (ICD-10 code F32) or recurrent depressive episode (ICD-10 code F33).

Exclusion criteria

The following individuals will be excluded from the study (Figure 2): 1) Patients with hospital contact up to 15 years prior to the first-time diagnosis with MDD due to the following A-diagnosis disorders: Manic episodes (F30), Bipolar affective disorder (F31), Schizophrenia (F20-F29), and Dementia (F00-F03), 2) Individuals with no available data on gender or date of birth from the Danish Civil Registration System (CRS)¹² at any time during the study inclusion period from 20011-2015, 3) Patients that have migrated to or from Denmark within the period of 15 years prior to the first-time MDD diagnosis, and 4) MDD patients under the age of 10 at the index date.

Figure 2. Flow chart of study sample inclusion and exclusion criteria.

Sample size calculation

To make the outcome significant, it is suggested to include at least 300 individuals in a Latent Class Analysis (LCA).¹³ We expect the number of included individuals in our study to exceed 300 considering an inclusion period of 5 years and a high prevalence of MDD.

Drugs for somatic disorders

The prescription of pharmacological treatment for somatic disorders will be available through the Danish National Prescription Register (NPR).¹⁴ The given somatic drug type is characterized by the Anatomical Therapeutic Chemical code (ATC code). All ATC codes except N05 and N06 (psychotropic drugs) will be defined as somatic drugs. For the purpose of not getting an excessive amount of observations, the drug types used in this study will be at the chemical subgroup level of the drugs (4^th level). A prescription drug will be considered as a taken drug if a patient has redeemed the drug at the community pharmacy at least twice during any of the 7-month intervals. The chemical subgroup drug types for somatic diseases are chosen based on the frequency of prescription: For each of the four intervals, prescription drugs redeemed by at least 0.5% of the study population will be considered; the combination of these ATC codes across all intervals will comprise the final drug portfolio to be submitted to the LCA model (Figure 3).

Figure 3. Process of determining the included ATC codes on 4^th level for all time-intervals.

Descriptive variables

The covariates used for this study are gender (male/female), age at index date (<18, 18-34, 35-64, and 65+), level of MDD severity, and education level. Gender and age are retrieved from the CRS. MDD symptom severity is categorized either as mild depression (F32.0, F33.0), moderate depression (F32.1, F33.1), severe depression (F32.2, F32.3, F33.2, F33.3), or unspecified (F32.4, F32.5, F33.4, F32.8, F32.9, F33.8, F33.9). This information is available through the DNPR. The educational level at index date will be available through the Danish Education Register (DER)¹⁵ and will be divided into four categories: ‘Primary lower secondary’, ‘Upper secondary’, ‘Bachelor, Masters, Doctoral’, or ‘Missing or not classified’.

Ethical considerations

This study is approved by the Regional Data Authorities (ref. P-2020-88). A Danish research study based on register data does not need ethical approval from the Danish Council on Ethics when the study does not involve biological material [The Danish National Committee On Health Research Ethics]. The data used for this study is obtained through Statistics Denmark (DST). All data is anonymized and solely aggregated data will be reported [DST Data for Research].

Latent Class Analysis

Latent Class Analyses (LCA) are used to identify unobservable subgroups (classes or clusters) within a population. These unobservable subgroups, called latent classes, have similar patterns of their shared set of observable variables from a data set. In this study, LCA will be performed utilizing the drugs that the MDD patients have redeemed at the pharmacy as indicator variables. The variables will be dichotomized into two categories: prescription drugs redeemed twice during an interval (1) and redeemed once or less (0). The latent variable will represent the drug profiles for somatic diseases in patients with MDD (Figure 4A).

Figure 4. A) Similarities in indicator variables (observation) determine the latent variables (drug profiles). B) Example of a transition in drug profile over time.

First, the study population will be divided randomly into two equal groups: a development cohort and a validation cohort. For the four respective time intervals, the LCA model will be fitted with a different number of classes for the development and validation cohort. This will be done to confirm that the best model fit for the development cohort corresponds to the best model fit for the validation cohort. The optimum number of latent classes will be determined by the Bayesian Information Criterion (BIC),^16,17 together with a clinical judgment of a psychiatrist. A decrease in BIC will indicate a better model fit, and the model that will yield the lowest BIC will be chosen.^18–20 In a case where the model fit for the development- and validation cohorts lead to a different number of latent classes, the clinical judgment will define the final number of classes.

After fitting the LCA model, the MDD patients will be assigned to a latent class (i.e. drug profile) where they have the highest probability of belonging compared to other classes (modal assignment). Similar analyses will be performed for each time interval. Each latent class/drug profile will have a unique pattern of redeemed prescription drugs based on the estimated item-response probabilities, i.e. the probabilities of receiving the chemical subgroup drugs given a particular class membership. All profiles will be characterized with the covariables gender, age, education level and MDD severity.

Treatment trajectories

The treatment trajectories are the transitions in latent class membership (drug profiles) over time (Figure 4B). Through modal assignment, all patients will be assigned to a class at each time interval. The trajectories will be identified by cross tabulating these assignments in each time interval. Thereby, the frequencies of each treatment trajectory will be obtained and thereafter, covariables will be associated with each trajectory. The trajectories will be visualized through a Sankey Diagram.

Study dissemination

Based on this study and the obtained data we will generate a research article which will submitted for publication to a peer-reviewed journal.