Population-wide analysis of differences in disease progression patterns in men and women

David Westergaard, Pope Moseley, Freja Karuna Hemmingsen Sørup, Pierre Baldi, Søren BrunakNature Communications 2019 · DOI

TL;DR

This study presents a comprehensive, population-wide analysis of disease progression patterns in men and women using Danish health registry data. It reveals significant sex-specific differences in disease incidence, age of diagnosis, and, crucially, the temporal sequencing of disease co-occurrences, highlighting the need for sex-stratified approaches in medicine.

What Problem Does This Paper Address?

Despite the recognized importance of sex and gender in health, sex-stratified medicine remains understudied. Much of the existing research focuses on single diseases or specific sexes, lacking a systematic, population-level understanding of how disease progression and co-occurrence differ between men and women. This gap hinders the development of truly personalized medicine, as it overlooks potential sex-based variations in disease pathways, risk factors, and temporal dynamics. The authors aim to address this by providing a data-driven model for analyzing these longitudinal differences across a broad spectrum of diseases.

Methods

The study employs a retrospective cohort design using the complete Danish National Patient Registry (NPR). Key methods include:

  • Incidence Rate Analysis: Calculating age-adjusted incidence rates (AIR) for individual diagnoses, comparing men and women.
  • Age at First Diagnosis: Analyzing the age at which individuals receive their first diagnosis for various conditions.
  • Diagnosis Co-occurrence Analysis: Identifying pairs of diagnoses that occur together, and importantly, determining the temporal order (directionality) of these co-occurrences. This was done using a sophisticated Bayesian hierarchical model to estimate relative risks and directionality, accounting for confounding factors like age, admission type, and time of year.
  • Diagnosis Trajectory Analysis: Constructing multi-step disease pathways by linking sequential diagnoses to uncover complex longitudinal patterns. This involved identifying trajectories with at least four diagnoses and a minimum of 100 patients.
  • Dual Terminology Use: Employing both the ICD-10 (International Classification of Diseases, 10th Revision) and Global Burden of Disease (GBD) terminologies to ensure robustness and capture different classification perspectives.

Datasets

  • Danish National Patient Registry (NPR): This dataset contains all public and private hospital admissions in Denmark from 1994 to 2015, covering approximately 6.9 million patients. It includes inpatient, outpatient, and emergency department contacts.
  • Global Burden of Disease (GBD) Terminologies: Used for a complementary analysis to validate findings and explore differences in classification.

Key Insights and Ideas

  • Sex as a Fundamental Risk Factor: The study strongly reinforces that sex is a significant, independent risk factor for disease, influencing not just incidence but also the very sequence and timing of disease development.
  • Beyond Anatomy and Genetics: While anatomical and genomic differences are known to play a role, many observed sex-specific incongruities in disease progression appear to be linked to socio-economic and ethological factors, suggesting a complex interplay of influences.
  • Temporal Dynamics Matter: The temporal order of disease co-occurrences is not arbitrary and can reveal crucial differences in disease pathways between sexes. What might appear as a random association in a cross-sectional view can become a distinct pattern when analyzed longitudinally.
  • Men's Trajectories Often Drive Patterns: The analysis of jointly observed longitudinal patterns was found to be most strongly driven by men, with a weaker strength of directionality observed in women, irrespective of the terminology used. This suggests that men's disease progression might follow more consistent, observable pathways in this dataset.
  • Underdiagnosis and Sex Differences: The study highlights potential sex-specific underdiagnosis, particularly in conditions like obstructive lung disease and osteoporosis, which can obscure true sex differences in risk and progression.

Key Results

  • Incidence and Age Differences: Over half of the ICD-10 diagnoses examined showed different incidence rates between men and women. Men were typically diagnosed at a younger age than women for most conditions, with notable exceptions like neoplasms and genitourinary system diseases.
  • Unique Co-occurrences: A substantial number of diagnosis co-occurrences (4155 directional pairs) were found to be unique to either men or women, indicating sex-specific disease relationships.
  • Reversed Temporal Relationships: The study identified instances where the temporal order of disease co-occurrence was reversed between men and women. For example, while ischemic heart disease precedes paroxysmal tachycardia in the general population and in men, the reverse was observed in women.
  • Trajectory Variations: Diagnosis trajectory analysis revealed significant disparities in patterns related to cancers, injuries, and substance abuse. For instance, injuries followed by alcoholic liver disease showed an increased relative risk in women, and complications following esophageal varices were also more prevalent in women. Obstructive lung diseases were found to be a risk factor for osteoporosis primarily in men, while osteoporosis without fracture followed by osteoporosis with fracture was observed only in women.

Limitations and Future Work

  • Data Limitations: The study relies on hospital admission data, excluding general practitioner visits and private specialist clinics, which might miss earlier disease stages or milder conditions. The quality of diagnosis recording, while generally high, cannot be entirely ruled out as a source of error. Recurrences of diagnoses were excluded to avoid bias.
  • Confounding Factors: While efforts were made to control for confounding factors like age, admission type, and time, latent factors and systematic errors in medical evaluation (under/overdiagnosis) could still influence results. Menopause, a significant factor for women, was not directly accounted for due to its absence in registry data.
  • Future Directions: The authors suggest extending this research to other cohorts (e.g., UK Biobank) and integrating genetic data (e.g., FinnGen, AllofUs) to identify genetic variants that might explain observed discrepancies. Further investigation into treatment-provoked co-occurrences and the role of hormonal influences is also warranted. The study also points to the need for better disease surveillance, diagnosis, and management strategies that are sex-stratified.

Summary updated

Abstract

Sex-stratified medicine is a fundamentally important, yet understudied, facet of modern medical care. A data-driven model for how to systematically analyze population-wide, longitudinal differences in hospital admissions between men and women is needed. Here, we demonstrate a systematic analysis of all diseases and disease co-occurrences in the complete Danish population using the ICD-10 and Global Burden of Disease terminologies. Incidence rates of single diagnoses are different for men and women in most cases. The age at first diagnosis is typically lower for men, compared to women. Men and women share many disease co-occurrences. However, many sex-associated incongruities not linked directly to anatomical or genomic differences are also found. Analysis of multi-step trajectories uncover differences in longitudinal patterns, for example concerning injuries and substance abuse, cancer, and osteoporosis. The results point towards the need for an increased focus on sex-stratified medicine to elucidate the origins of the socio-economic and ethological differences.
BibTeX
@inproceedings{westergaard2019population,
  author = {David Westergaard and Pope Moseley and Freja Karuna Hemmingsen Sørup and Pierre Baldi and Søren Brunak},
  title = {{Population-wide analysis of differences in disease progression patterns in men and women}},
  booktitle = {Nature Communications},
  year = {2019},
  doi = {10.1038/s41467-019-08475-9},
}
Figure