Ready for Winter? Seasonal Pneumonia Burden and Health-System Capacity

The study covers Lebanon against its Eastern Mediterranean neighbours from 1990 to 2023, built around five sequential questions: how big is the burden and is it improving, when does it happen, where and to whom, why, and what happens next.

Each question maps to one view of a six-view Power BI dashboard and to one of four analytics modes — descriptive, diagnostic, predictive, prescriptive.

• Three public sources, each used for what it measures best: IHME Global Burden of Disease 2023 for burden and attributable risk, WHO FluNet for weekly seasonality across 31 seasons, WHO Global Health Observatory for immunisation coverage.
• Python (pandas) for cleaning and reshaping the raw exports; a star schema in Power BI for the model.
• Age-standardisation to WHO world-standard weights, estimated annual percentage change from a log-linear regression with confidence intervals, seasonal decomposition of the weekly series, population attributable fractions for 1990 versus 2023, and an OLS extrapolation of the log-rate trend for a ten-year forecast.

Lebanon's age-standardised LRI mortality fell about 1.56% a year to 39.4 per 100,000 in 2023 — then the decline stalled after 2019 and partly reversed.

The regional influenza series peaks reliably in week 7, mid-February, with a median nine-week gap between the alert threshold and the peak. That gap is the operational finding: it is the window in which beds, oxygen and staffing can be moved. Lebanon's own surveillance series was too fragmented to analyse independently, which is itself a finding about surveillance capacity rather than a gap in the analysis.

The risk profile has inverted since 1990. Child-nutrition risks collapsed while ambient particulate matter (17.6%) and tobacco (16.8%, roughly tripled) moved to the top. Meanwhile childhood DTP3 coverage fell from about 83% before 2019 to below 50%. The burden is concentrating in exposed older adults exactly as preventive capacity contracts.

• GBD figures are modelled estimates, not counted events, and in a data-scarce setting they carry real uncertainty. The forecast inherits every one of those assumptions.
• Influenza surveillance is a proxy for pneumonia seasonality, not a measure of it.
• The analysis is ecological: associations hold at population level and say nothing about individuals.
• One result ran against expectation — a high-income Gulf state topping the mortality ranking. We verified it against the source and reported the uncertainty rather than dropping it. Keeping that in the deliverable was the right call and I would do it again.