The relationship between outdoor air pollution and respiratory health outcomes is one of the most thoroughly documented associations in environmental epidemiology. Dozens of studies have established the link between short-term PM2.5 spikes and increased emergency department visits, hospital admissions, and mortality. Yet most hospital systems still operate without any systematic air quality forecasting integrated into their capacity planning workflows. This gap represents both a clinical failure and an operational inefficiency that costs the healthcare system billions of dollars annually.
A collaborative study conducted by Trace AQ and researchers at the University of Texas Dell Medical School sought to quantify what hospitals could gain from incorporating hyperlocal, 48-hour PM2.5 forecasts into their operational planning. The results were striking enough that we felt compelled to share them broadly.
Study Design
We analyzed emergency department admission records from four Austin-area hospitals over a three-year period from January 2022 through December 2024. We focused specifically on admissions with primary diagnosis codes associated with respiratory conditions: asthma exacerbations, COPD exacerbations, acute bronchitis, pneumonia, and non-specific chest pain with documented respiratory symptoms.
For each day in the study period, we retrospectively ran our Trace AQ forecasting model to generate what the 24-hour, 48-hour, and 72-hour forecasts would have predicted for PM2.5 concentrations at each hospital location. We then correlated these forecast values with actual same-day and next-day respiratory admission counts.
The key methodological choice was to evaluate forecast accuracy as it would have existed operationally — including all the uncertainties and errors a real-time forecast would have carried. This is not a post-hoc correlation between observed air quality and admissions, a well-established relationship. We wanted to know whether forecast air quality is predictive of future admissions in a way that could actually inform hospital staffing decisions made 24–48 hours in advance.
Key Findings
At 24-hour lead times, PM2.5 forecast quintile was significantly predictive of next-day respiratory admissions. Hospitals in the highest quintile of forecast PM2.5 experienced 34% more respiratory admissions on average than hospitals in the lowest quintile, controlling for day of week, season, and baseline respiratory illness prevalence.
At 48-hour lead times, the predictive relationship weakened but remained statistically significant and operationally meaningful. Forecast PM2.5 in the top quintile at 48 hours was associated with a 23% elevation in respiratory admissions — our headline finding, consistent with what we've reported in prior summaries of this work.
The practical implication is significant. A hospital administrator who sees a Trace AQ 48-hour forecast showing elevated PM2.5 — say, above 35 µg/m³ — can use that information to adjust respiratory therapy staffing, ensure adequate nebulizer and supplemental oxygen supply, and reduce same-day elective procedure bookings that would compete for bed capacity. These are decisions that can be made two days in advance, and the data strongly supports making them.
The Geographic Precision Advantage
One finding that surprised even our research team was how much spatial precision mattered. Austin is a geographically varied city — the western hills trap pollutants differently than the eastern flatlands, and industrial corridor proximity drives local NO2 and PM2.5 elevations that county-level monitoring doesn't capture.
When we compared predictions based on the nearest EPA monitoring station (the traditional approach) against Trace AQ's hyperlocal grid, the hyperlocal approach improved 48-hour admission prediction accuracy by an additional 18% at the two hospitals located farthest from EPA monitoring stations. For healthcare systems in rural areas with sparse monitoring networks, this advantage would be even larger.
Why Hospitals Haven't Acted Yet
Given the strength of the evidence, it's fair to ask why hospitals aren't already integrating air quality forecasting into clinical operations. The honest answer is threefold. First, the data simply wasn't accessible in a clean, API-driven format that hospital IT systems could consume. Most air quality data exists in government portals designed for researchers, not operational integration. Second, the forecasting quality historically wasn't good enough to be actionable — if a forecast is wrong 40% of the time, it creates as many problems as it solves. Third, the clinical workflow integration hadn't been mapped out by anyone willing to do the unglamorous implementation work.
Trace AQ addresses the first two problems directly. Our REST API delivers hourly PM2.5 forecasts in a format any healthcare IT system can parse. And our physics-constrained architecture, as documented in our companion technical article, delivers forecast accuracy that meets the bar for operational use. The third challenge — workflow integration — is something we're actively tackling through pilot programs with health system partners.
Beyond Emergency Departments
The study focused on emergency admissions because the data is cleanest and the operational implications are most direct. But the implications of air quality forecasting for healthcare extend much further. Pulmonologists can use the data to proactively contact high-risk patients before air quality events and adjust medication regimens. Chronic disease management programs can correlate symptom trajectories with air quality exposure. Pediatric asthma programs — which serve one of the most vulnerable populations — can use school-level forecasts to trigger preventive protocols.
Each of these applications requires the same underlying infrastructure: reliable, hyperlocal, multi-day air quality forecasts delivered through an accessible API. Building that infrastructure is what Trace AQ does. The clinical applications are what our healthcare partners bring to the table.
We're actively expanding our health system pilot program and welcome inquiries from hospital systems, public health agencies, and clinical research teams interested in incorporating air quality data into their work.
