Community clozapine, ECT, and lithium use are inversely associated with excess suicide rates in adolescent males

Data sources and initial processing

Data published by the Swedish National Board of Health and Welfare (available in Swedish: https://sdb.socialstyrelsen.se/if_dor/val.aspx; https://sdb.socialstyrelsen.se/if_lak/val.aspx)^26,27 From 2016 to 2020, 21 Swedish counties with age ranges 15-19 and 20-24, respectively, were sampled for the following variables: Confirmed suicide rates per 1000 population (based on estimated population from 1 January of the year recorded) and per 100,000 population for lithium (ATC code N05AN01) and clozapine (ATC – code N05AH02). All data were extracted for both males and females, for each sex, across all Swedish counties from 2016 to 2020. Similarly, published data from the Swedish ECT Registry (available in Swedish: https://ect.registercentrum.se/statistik/utdata-ect/p/rJ3vhF3Lw)²⁸ Details of the number of patients (age group and gender) receiving ECT treatment were extracted for each region from 2016 to 2020 and age range 0–17 (<18). All individuals who received ECT treatment were aged 13 years or older, and the overwhelming majority of the sample consisted of individuals aged 15–17 years. Reasons for including 2016–2020 and the variables included can be found in Supplementary Materials. is explained in detail. Source data, dimensions, and other important information are also available here to facilitate proper replication of results.

Statistical considerations

Several measures have been taken with the aim of enhancing the robustness of the included variables. These initial steps were performed using Microsoft Excel for Microsoft 365 MSO (version 2204 build 16.0.15128.20278) and are detailed in the supplementary material.

min-max normalization to account for both baseline magnitude and variability in regional annual data

In short, the mean and median for each region and gender group from 2016 to 2020 exhibited larger standard deviations than the mean/median for most of the key variables, so the underlying data distribution was considered not representative. Therefore, normalization of minimum and maximum values ​​was implemented across each year, region, and treatment type (i.e., DerivedValue, etc.).₂₀₁₆= value₂₀₁₆-min(value_2016-2020)/(Maximum value_2016-2020)-minutes(value_2016-2020))). For suicide rates, the difference between adolescent and young adult suicide rates per 100,000 inhabitants was calculated by region, year, and gender group (i.e., DerivedExcessSuicideRateAdolescence₂₀₁₆= suicide rate adolescents₂₀₁₆–Suicide Rate Young Adulthood₂₀₁₆)) then undergoes min-max normalization. The average minimum-maximum value normalized from 2016 to 2020 was implemented in subsequent analyses, and each region and gender group was compared to the mean treatment value measured from 2016 to 2020 and excess adolescent suicide. Represented by one value of death. When compared to the median of the normalized values ​​of the minimum and maximum values, for example, counties providing treatment only for his 1 or 2 years in his 5 years under study are also recognized, so that the mean is It was considered to be more representative of the underlying data distribution. Validity of the min-max normalized value to explain the baseline magnitude of treatment use frequency as well as up-regulating regions that show low year-to-year variability—the derived treatment proxy variable Baseline values ​​of clozapine, ECT, and lithium investigated by Pearson correlation for yearly sums – show moderate to strong positive correlations (r= 0.49).

Subtract suicide rates within regions and between age groups to reduce estimated confounding

To reduce the impact of potential unmeasured causes of confounding on suicide rates across the county, the primary outcome variable was based on subtracting adolescent and young adult suicide rates. Thus, the impact of regional differences on suicide rates, such as in population size, socioeconomic status, substance abuse, or the availability and quality of psychiatric care, may thus be significantly reduced. Similarly, subsequent normalization of minimum and maximum values ​​contributed to the normalization of the potential confounding effect of substantial outliers on baseline absolute values.

Input variable considerations

Basing lithium and clozapine values ​​on dosing frequency reduced the effects of short-term treatment, and normalization of minimum and maximum values ​​showed very high values ​​for 1 or 2 years and other values. year contributed to downregulation of potential confounders from regions showing zero (no treatment). Considered reconcilable with best practice care.

Rationale for selected study period

The duration was chosen to be consistent with the direction of the major courses of national treatment guidelines for ECT since 2016, postpubertal with mood-matching psychotic symptoms, catatonia, or severe MDD with treatment. It is recommended for use in the care of adolescents with the highest priority. resistance. Similarly, treatment frequency for 2021–2022 was not included in the analysis, as information on local suicide rates was available through 2020 at the time the data were available.

Consolidate frequency of clozapine, ECT, and lithium use

Implementing average treatment values ​​for frequency of clozapine, ECT, and lithium use would further improve robustness while reducing ECT aversion (or clozapine or lithium aversion) allows for county recognition.

Implementing a Weighted Robust Regression Model to Further Reduce Confounding

The use of robust linear regression models weighted on regional population estimates contributed to reducing confounding from outlier counties with low populations (which allowed a single clinician influence to reduce the measured We would expect the frequency of visits to vary significantly, but may not be representative of the overall quality of care provided (regional).

statistical analysis

Initial data processing (calculation of min-max normalized values) was performed using Microsoft Excel365 MSO. [Version 2210 Build 16.0.15726.20188] 64-bit). All downstream statistical analyzes were performed using R version 4.0.3. It is detailed in the supplementary material. In summary, all variables were investigated for normal distribution.Variables not meeting the requirement for normal distribution were transformed with Blom’s method²⁹ For subsequent robust linear regression analysis (except for nonparametric tests, i.e. exact Wilcoxon rank sum tests). Exploring potential disruptions resulting from his Covid-19 pandemic outbreak in 2020^{twenty four}, implemented the Wilcoxon Rank Sum Exact test to compare the mean normalized values ​​of the minimum and maximum excess adolescent suicide rates from 2016 to 2019 to the 2020 and 2020 values ​​across the 21 regions. compared. sex: p-value = 0.811, Female: p-value = 0.613, Male: p-value = 0.358). Multiple ordinary least-squares regressions were implemented to examine the relative contribution of each treatment to the mean clozapine/ECT/lithium surrogates.Primary analyzes of associations between regional adolescent excess suicide rates and mean frequency of use of clozapine/ECT/lithium treatments were investigated by robust linear regression models using the R package ‘robustbase’³⁰recommended settings (KS2014), and a standard MM regression estimator that guarantees an acceptable compromise between high breakdown (i.e. 50%) and very high efficiency (i.e. 95%).³¹) and adjusted weights (proportional to 2020 regional population and national estimates for the same year)³². exhibition model p– A value of <0.05 for the main explanatory variable was considered significant. The main models are shown in x–y scatterplots, showing the slope coefficients estimated from the regression model (Figure 1). 1). To validate the significant model, a one-tailed Wilcoxon rank sum exact test was used to find that counties in the lower quartile (Q1) of adolescent excess suicide rates were more likely to The hypothesis was tested whether the use of advanced treatments is high in P.– Values ​​< 0.05 were considered significant. A post-hoc analysis was then performed to determine associations in gender and treatment-stratified groups (i.e., men and women, clozapine, ECT, and lithium). Treatment variables were investigated for collinearity and such associations were adjusted in downstream analyses. To reduce the possibility of bias due to model overfitting (because of the small sample size, n= 21 regions), treatment variables were tested separately for the female subgroup, and significant values ​​of results were subjected to strict Bonferroni correction³³Similar to the main analysis, significant associations were verified by contrasting the dichotomized 25th quartile based on excess suicide mortality with candidate treatment variables. In a post hoc analysis of these non-Blom-transformed values, p– Values ​​<0.05 (Bonferroni-adjusted for females) were considered significant.

Report overview

For more information on the study design, please visit Overview of the Nature Portfolio Report Link to this article.