Development of an algorithm for finding pertussis episodes in a population-based electronic health record database

Area of Science:

• Infectious Diseases
• Machine Learning in Healthcare
• Public Health

Background:

• Pertussis (whooping cough) burden in adolescents and adults is underestimated due to atypical presentations.
• Current vaccination strategies for tetanus-diphtheria-acellular pertussis (Tdap) may be suboptimal for this demographic.
• Atypical pertussis presentation hinders accurate diagnosis and effective control measures.

Purpose of the Study:

• To develop and validate a machine learning algorithm for identifying undiagnosed or misdiagnosed pertussis in patients with acute respiratory disease (ARD).
• To leverage electronic health records, including clinician notes and demographic data, for improved pertussis case identification.
• To enhance understanding of clinical indicators associated with pertussis in adult and adolescent populations.

Main Methods:

• A machine learning model (LightGBM) was developed using electronic health records from 2007-2019.
• Two cohorts were used: positive pertussis (4,515 episodes) and negative pertussis/ARD (4,573,445 episodes), with a focus on episodes having ≥7 symptoms.
• Model performance was evaluated using laboratory-confirmed cases and explainability was assessed with Shapley additive explanations.

Main Results:

• The algorithm demonstrated a recall of 0.72 and specificity of 0.94 in identifying pertussis episodes among ARD diagnoses.
• When applied to laboratory-confirmed cases, the model achieved a specificity of 0.846.
• Clinical notes indicating whooping cough, whoop, and post-tussive vomiting increased the predictive probability of pertussis.

Conclusions:

• Machine learning offers a viable approach to identify potential pertussis cases within broader acute respiratory disease diagnoses.
• The algorithm can aid clinicians in recognizing atypical pertussis presentations in adolescents and adults.
• Improved identification of pertussis can inform public health interventions and optimize vaccination efforts.