Clinical Scorecard: Leveraging Community Syndromic Surveillance to Predict School Closures Due to Enterovirus: A Real-Time LSTM Method

At a Glance

Key Highlights

• Enterovirus infections cause recurrent pediatric epidemics, particularly in East and Southeast Asia.

• Evidence cited by the study indicated that enterovirus infections in Taiwan increased by 619.5% after declining during the COVID-19 pandemic.

• The analysis included daily outpatient data from 110 participating clinics from January 2022 through January 2025.

• District-specific LSTM models used a 30-day lookback window to forecast EV-like syndromic activity up to 14 days in advance.

• The framework achieved accuracies of 0.94 for preschool data and 0.96 for primary school data.

• Model-generated alarms generally occurred before abnormal class suspension days.

Guideline-Based Recommendations

This retrospective modeling study did not establish clinical guidelines or recommend its framework for routine implementation without further validation.

Diagnosis

• The EV-like illness surveillance category was defined using ICD-10 codes for HFMD, HA, and other enterovirus-related diagnoses.

• HFMD and HA were also analyzed as separate syndrome groups.

• These syndrome groups served as surveillance indicators and were not presented as new diagnostic criteria.

• The study analyzed aggregated records without patient-level identifiers.

Management

• The study did not evaluate clinical treatment or outbreak-control interventions.

• Model-generated alarms may support timely local responses, targeted class suspension decisions, and preparedness among schools and families.

• The framework requires further validation before its public health applicability can be established in other settings.

Monitoring & Follow-up

• Daily pediatric outpatient data were aligned with school class suspension records and total class counts at the district level.

• Abnormal suspension days were defined using district-specific suspension-rate thresholds.

• LSTM residuals were evaluated using district-specific adaptive thresholds to account for differences in baseline activity and variability.

• The study did not establish a patient-level monitoring or follow-up protocol.

Risks

• Enterovirus infections spread rapidly among children and may result in preschool or primary school class suspensions.

• In the preschool district-level analysis reported in the main text, false-positive alarms occurred in some districts, but no false negatives were observed under the study’s lead-time matching definition.

• Model performance may be affected by missing data, data quality, or external changes not represented in historical observations.

• Associations between EV-like activity and suspensions were assessed only at the aggregate district level.

Patient & Prescribing Data

The study examined syndromic data for children aged 0–6 years and 7–12 years across Taipei City. Records were aggregated by date and district and contained no patient-level identifiers.

The study did not evaluate medications, prescribing patterns, or treatment effects.

Clinical Best Practices

• Interpret ICD-10–based syndrome groups as surveillance indicators rather than laboratory-confirmed infections.

• Use district-specific calibration to account for geographic differences in activity and variability.

• Evaluate model alarms alongside observed outpatient activity and school suspension data.

• Recognize that alarms indicate unusual syndromic activity rather than confirmation of an outbreak or prediction of suspensions at individual schools or classes.

• Validate the framework in additional settings and with meteorological, mobility, environmental, or virological predictors.

Related Resources & Content

1. Using Community Syndromic Surveillance to Anticipate Enterovirus-Related School Class Suspensions: A Real-Time LSTM Approach — Rayguru C, Jian HL, Peng YF, et al. International Journal of Infectious Diseases. 2026;170:108918. doi:10.1016/j.ijid.2026.108918.

2. Aggregated EV-Like Illness Data Analyzed in the Study — Available through Figshare. doi:10.6084/m9.figshare.32017062.

3. Software Used for the Study’s LSTM Surveillance Framework — Available through Figshare. doi:10.6084/m9.figshare.32006418.