Estimation of Suicide Relative Risk at the Municipal Level Using Administrative Health Records and Count Regression Models
12/11/2026 –, Multifuncional (Remoto)

Introduction

The spatial scale problem in official statistics

Geographic scale plays a fundamental role in spatial epidemiology because it determines the ability to identify territorial disease patterns and support public health interventions [1], [2]. When health information is reported at aggregated spatial levels, important local variations may remain hidden, limiting the identification of municipalities where suicide risk differs substantially from expected levels [3].

In Mexico, the National Institute of Statistics and Geography (INEGI) publishes official suicide statistics primarily as age-standardized mortality rates at the state level [4]. Although these statistics describe the overall pattern of suicide and allow comparisons among states, they do not capture local variations in suicide risk or incorporate information on mental health conditions and territorial characteristics.

Consequently, state-level indicators may provide an incomplete picture of the phenomenon. States with relatively low suicide rates, such as Guerrero, Oaxaca, or the State of Mexico, may still contain municipalities with substantially elevated risks. Conversely, high state-level rates do not necessarily imply a homogeneous spatial distribution, since excess risk may be concentrated in only a few municipalities. This aggregation limits the identification of priority municipalities for territorial analysis and suicide prevention.

Research gap

Relative risk estimation is widely used in epidemiology to identify areas where disease occurrence is higher or lower than expected after accounting for demographic structure [5]. Count regression models have been successfully applied to estimate disease risk at small geographical scales [6].

However, our literature review did not identify published studies proposing a methodology to estimate municipal suicide relative risk in Mexico by integrating administrative health records, demographic information, and territorial indicators within a count regression framework. One reason is the absence of municipal-level prevalence estimates for mental disorders.

To overcome this limitation, this study uses administrative healthcare records as indirect indicators of population mental health conditions. Although these records do not measure prevalence directly, they provide a continuous, standardized, and geographically referenced source of information.

The objective of this work is therefore to develop a reproducible methodology for estimating municipal suicide relative risk using administrative health records, demographic information, and territorial indicators.

Methodological framework

The proposed methodology estimates municipal suicide relative risk for the period 2013–2023 through four stages.

Stage 1. Expected cases. Expected suicides were estimated through indirect standardization [5]. National age- and sex-specific suicide rates were computed as

$
$r_{jt}=
\frac{\sum_i y_{ijt}}
{\sum_i n_{ijt}}
$$

and expected cases as

$$
E_{it}=\sum_j r_{jt}n_{ijt},
$$

where $E_{it}$ denotes the expected number of suicides in municipality $i$ during year $t$. Expected cases constitute the epidemiological baseline and were incorporated into the regression model as an exposure (offset) term.

Stage 2. Analytical dataset. A municipality–year database was built by integrating observed suicides, expected cases, administrative healthcare records compiled by the National Institute of Psychiatry Ramón de la Fuente Muñiz, and territorial indicators from CONAPO. Administrative records included depression, legal and illegal substance use, domestic violence, and the availability of psychologists and psychiatrists. The final dataset comprised 1,399 municipality–year observations from 198 municipalities across Mexico.

Stage 3. Variable selection. Lasso regularization for count regression models was applied to identify informative explanatory variables while reducing multicollinearity [7].

Stage 4. Relative risk estimation. Owing to overdispersion in suicide counts, a Negative Binomial regression model was fitted [8], [9]:

$$
Y_i\sim NB(\mu_i,\theta)
$$

with

$$
\log(\mu_i)=\log(E_i)+\mathbf{x}_i^\top\boldsymbol{\beta},
$$

where $\log(E_i)$ is the offset term. Municipal relative risk was estimated as

$$
RR_i=\exp(\mathbf{x}_i^\top\boldsymbol{\beta}),
$$

where $RR_i>1$ indicates higher-than-expected risk and $RR_i<1$ lower-than-expected risk.

Main findings

The proposed methodology estimated municipal suicide relative risk for 198 municipalities and generated annual risk maps for 2013–2023. Results revealed substantial local spatial variation that remains hidden when analyses rely exclusively on state-level statistics.

The final model identified psychiatrist availability as a protective factor (IRR = 0.774), whereas psychologist availability (IRR = 1.765), illegal substance use (IRR = 1.138), domestic violence (IRR = 1.128), households with dirt floors (IRR = 1.246), and households without drainage (IRR = 1.561) were associated with increased relative risk.

As an illustration, Acapulco de Juárez exhibited a relative risk approximately four times higher than expected at the beginning of the study period and more than nine times higher than expected in 2023, despite belonging to Guerrero, one of the Mexican states with the lowest official suicide mortality rates.

Contributions

This work develops a reproducible methodology for estimating municipal suicide relative risk by integrating indirectly standardized expected cases, administrative healthcare records as indirect indicators of mental health conditions, and territorial indicators within a Negative Binomial regression framework. To the best of our knowledge, our literature review did not identify published studies proposing a comparable methodological framework for municipal suicide risk estimation in Mexico.

The methodology complements official statistics by producing municipal-level estimates and maps of suicide relative risk, allowing local spatial variations to be identified that remain hidden in state-level analyses. The proposed framework provides a reproducible approach for territorial analyses of suicide and for identifying municipalities with elevated relative risk.

References

[1] Elliott et al., Spatial Epidemiology, 2000.

[2] Waller and Gotway, Applied Spatial Statistics for Public Health Data, 2004.

[3] Openshaw, The Modifiable Areal Unit Problem, 1984.

[4] INEGI, Suicide Statistics, 2025.

[5] Moraga, Geospatial Health Data, 2023.

[6] Cameron and Trivedi, Regression Analysis of Count Data, 2013.

[7] Wang et al., Statistical Methods in Medical Research, 2016.

[8] McCullagh and Nelder, Generalized Linear Models, 1989.

[9] Hilbe, Negative Binomial Regression, 2011.