Health and Social Services During Heat Events: Demand for Services in Los Angeles County

Issue

Local governments are at the frontline to ameliorate the human and social impact of climate change. Recent local data on the relationship between climate factors and health and social outcomes are informative for developing effective operation plans and assessing outcomes. In 2021, Los Angeles County published a Climate Vulnerability Assessment highlighting the potential risks from future climate change. The Climate Vulnerability Assessment did not draw from the county's own data about outcomes associated with heat and other climate risks. Routinely collected data from Los Angeles County may provide more timely and actionable information on the impact of extreme weather events. County departments do not yet have the tools or experience to assess the impact of extreme weather events on their operations or on their clients.

Research Objectives

The project has four objectives:

1. identify and obtain data routinely collected by Los Angeles County that could provide information about the impact of extreme heat on services and population outcomes
2. compare metrics to classify extreme heat events in terms of predictive power of adverse events, while considering ease of use, interpretability, and suitability for communication
3. quantify the association between extreme heat events and adverse outcomes
4. create a user-friendly software tool that allows county departments and other organizations to create updated results with their own data using a matching approach.

Methods

Outcome Measures

We explored a range of possible data sources, either directly from departments or through the Chief Information Office. This study describes results for emergency medical calls to the Los Angeles County Fire Department (LACoFD), unexpected deaths investigated by the Medical Examiner, emergency room visits collected for syndromic surveillance, and bookings by police and sheriff departments. Other data investigated include emergency room visits in three county hospitals, mental health crises, and births and premature births.

Heat Measure

In a preliminary analysis, we compared the predictive performance of several potential measures ranging from simple daily temperature thresholds to general weather patterns and indices with multiple inputs. Included among the more complex measures where the National Weather Service (NWS) HeatRisk Version 2.5, the Australian Excess Heat Factor (EHF), and the Spatial Synoptic Classification (SSC).

The NWS HeatRisk is a categorical index with five levels (none, minor, moderate, major, extreme). It is still under final development (as of June 2024) and has recently been upgraded from “prototype” to “experimental” status. The EHF is a scale used in Australia that combines short-term heat measures and acclimatization. The SSC is a classification of weather patterns in larger geographic areas and has been adopted for warning systems in various cities.

Analysis

Results are based on two different approaches: a regression approach using count models and a matching approach. Regression approaches using count models are commonly used in research publications examining the effect of heat on health outcomes. We also used regression models with split-sample cross-validation and mean squared prediction error in the preliminary analysis to compare different heat measures.

Regression models are efficient because they use information from all observed days, but they are vulnerable to biases from misspecification. Their complexity makes them less transparent and less practical outside research settings.

As an alternative, we use a matching approach that modifies a method initially developed by the California Department of Public Health. In the matching approach, heat event days are only compared to (matched) non-heat days within a narrow time window. Statistical estimates from matching are less precise than in (correctly specified) regression models because only a subset of all data is used, but they are more robust to misspecification. We create software for an online tool that allows users to easily analyze their data using the matching approach. It allows the user to pre-select the number of potential matched days between two and ten.

Key Findings

Heat Metric Selection

The NWS HeatRisk Version 2.5 performed best overall when comparing errors for out-of-sample prediction across different heat metrics and outcome variables in Los Angeles County. It is available at geographically granular resolution, with local seven-day forecasts available nationwide, and is interpretable to lay audiences.

For mortality data, the EHF performed slightly better than HeatRisk, but only as a continuous measure and only when calculated in a way modified from its original publication. The EHF scale has no intuitive interpretation, which makes it less useful for communication. There are no identified thresholds at which policymakers should take action. For Los Angeles County, the SSC had lower predictive power than HeatRisk for all outcomes. The SSC is a general weather pattern without smaller geographic resolution and can only be calculated for a limited set of stations with a proprietary algorithm.

There were no large differences in predictive power between heat measures and any heat measure accounts for a small fraction of the variation. Even a simple average of daily temperature or a three-day average is almost as predictive as the more complex measures (and, for mortality, outperforms the SSC).

The main results reported in this study are calculated using HeatRisk.

Deaths Investigated by the Los Angeles County Medical Examiner

Compared to days with no HeatRisk, days with moderate, major, or extreme HeatRisk were associated with increases in deaths investigated by the Medical Examiner of 6.7% (95% confidence interval [CI]: 1.9–11.7%), 15.3% (CI: 2.9–29.1%), and 65.5% (CI: 34.9–102.1%), respectively (Figure 1). These effects were more pronounced for individuals who were homeless or in care homes (including care homes, nursing homes, retirement homes, or convalescent homes). Relative to days with no HeatRisk, major or extreme heat days were associated with a 59.3% (CI: 19.8–109.4%) increase in deaths among homeless individuals and a 91.4% (CI: 19.0–198.6%) increase in deaths among those in care. These results were based on a regression model for data from 2014–2019.

NOTE: Based on regression models, comparison is no HeatRisk. Values on the y-axis are multiplicative.

Bookings for Violent Offenses

Relative to days with no HeatRisk, days with minor, moderate, major, or extreme HeatRisk were associated with an increased number of bookings for violent offenses of 3.6% (CI: 2.7–4.5%), 5.7% (CI: 4.4–7.0%), 7.9% (CI: 4.6–11.3%), and 11.9% (CI: 6.6–17.5%). These estimates are shown in Figure 2. Translated to absolute numbers, HeatRisk was associated with an additional 1,442 bookings for violent offenses annually (approximately 2.6% of all such bookings in Los Angeles County).

NOTE: Based on regression models, comparison is no HeatRisk. Values on the y-axis are multiplicative.

Emergency Room Visits

Counts of emergency room visits come from syndromic surveillance data. Syndromic surveillance enables fast detection of emerging health issues, such as new infectious disease outbreaks. Most emergency rooms in Los Angeles County provide data to this system. We obtained aggregated counts by day, without information on location or causes, beginning in 2016. Applying the matching approach to the period of 2016–2023 with a three-week comparison window, major HeatRisk days (HeatRisk level 3 red) had an additional 188 visits and extreme HeatRisk days (HeatRisk level 4 magenta) had an additional 295 visits, but these point estimates have a wide CI, as shown in Figure 3. Figure 3 is created with the software tool developed by this project, which will enable users to create updated findings.

NOTE: Based on matching method, comparison line 0 is average of none (no HeatRisk) and minor HeatRisk. Federal holidays included for heat days. Values on the y-axis are additive.

Based on the 2021 Medical Expenditure Panel Survey (the latest year available), we calculate the average (nonzero) amount paid for an emergency room visit (insurance and copayments) to be $1,181. Applying the point estimate of the increase estimated for 2016–2023, the eight extreme heat days in 2022 led to additional emergency room expenses of $2.3 million.

Los Angeles County Fire Department Emergency Medical Services

The LACoFD provided information on emergency medical service (EMS), including a geographic breakdown and the cause of service calls. We show both regression and matching results for this outcome.

Using regression models and compared to days with no HeatRisk, Extreme HeatRisk was associated with an increase of 14.9% (CI: 12.8%–16.9%), Major HeatRisk produced an 11.6% increase (CI: 10.6%–12.6%), and Moderate and Minor HeatRisk produced 6.8% (CI: 6.3%–7.3%] and 2.8% (CI: 2.5%–3.1%) increases (Figure 4). Figure 4 in addition shows how these values differ across the populations served by different stations, using the Social Vulnerability Index developed for Los Angeles County's Climate Vulnerability Assessment. Point estimates for stations in areas with higher Social Vulnerability tend to be slightly larger, but these differences are not statistically significant.

The graphs are very similar for all four groups, with minor at about 2 or 3 percent, moderate at about 7 to 8 percent, major at about 12 to 13 percent, and extreme at about 15 to 17 percent. Extreme is highest for the unhoused group.

NOTE: Based on regression models, comparison is no HeatRisk.

Figure 5 displays results countywide using the matching approach and the software tool.

NOTE: Based on matching method, comparison line 0 is the average of none and minor HeatRisk. Values on the y-axis are additive.

Recommendations

We compared several competing indicators and found HeatRisk, a new measure developed by the NWS, well suited for planning and impact assessment. HeatRisk was more predictive than other categorical indicators for the outcomes investigated. HeatRisk can be calculated for small geographic areas, requires only temperature data, is predictive of adverse outcomes, is easy to communicate, and provides discrete thresholds for action.

In order to make HeatRisk more widely used in research and routine evaluations, the NWS may want to consider an Application Programming Interface that includes historic data by location (latitude/longitude). Currently, only forecasted future values are available. Combined with the software for the matching analysis available with this study, end users could quickly evaluate heat effects for their particular outcome and location.

Los Angeles County collects numerous types of data that could inform decision-making across departments. The 2024–2030 Strategic Plan of Los Angeles County emphasizes data-driven decisions, and facilitating data sharing across departments can ensure that policy and operational recommendations are well informed. Reducing hurdles to data sharing across departments and outside collaborators would contribute towards Focus Area E. An example is our finding that the number of unexpected deaths is especially elevated in the unhoused population and among older adults receiving assistance in their living situation. This new (though not unexpected) result was only possible because the Medical Examiner data was linked to other county department data. Knowing the vulnerability of this group of county clients may allow preventive outreach.

Heat events have a significant impact on Los Angeles County, from increased calls for EMS to the LACoFD, more emergency room visits, and a larger number of unexpected deaths investigated by the county Medical Examiner. It is not yet a comprehensive picture of the potential adverse effects of climate change in the county, but a first necessary step. There are many unknown risks, but quantifying those that can be quantified will improve planning.

The last chapter in the county's Climate Vulnerability Assessment highlighted the interaction between risks. The largest risks are most likely to result from interacting and simultaneous events. What are the risks and consequences when extreme heat combines with extreme air pollution or power failures? Interactions are beyond the scope of this study, which focuses on extreme heat only, but deserve more attention in future research.