Impact of Hospital Catchment Area Definitions on COVID-19 Forecasts

During the COVID-19 pandemic, short-term Forecasts played a crucial role in public health policy by helping manage hospital resources. A key concern was whether hospitals would have enough capacity to care for patients. To address this, defining hospital catchment areas became important for planning and understanding local healthcare needs.

#What are Hospital Catchment Areas?

Hospital catchment areas are regions that indicate where patients are likely to seek care. These areas help health officials estimate how many people will use hospital services and what kind of care they might need. Catchment areas are useful for budget planning and staff allocation. They also help identify if some regions have too many or too few hospital services. Finally, they allow for the calculation of admission rates and vaccination coverage.

#COVID-19 and Hospital Admissions

In England, COVID-19 cases and hospital admissions were reported at different administrative levels. COVID-19 cases were reported by Local Authorities, which are small regions, while hospital admissions were reported by NHS Trusts, which manage multiple hospitals. To accurately forecast hospital admissions, local COVID-19 cases needed to be defined within each hospital's catchment area.

However, catchment areas in England are not clearly defined because patients can choose where to go for care. NHS Trusts do not have set geographical boundaries, and there is no complete list of registered patients. A good catchment area definition should reflect where patients actually go for care and consider factors like travel distance and hospital size.

#Estimating Catchment Areas

Catchment areas can be estimated using hospital admission data or simple rules. If data on past admissions is available, it can help define these areas. Two common methods are:

1. First-past-the-post Model: This assigns each geographical area to the hospital where most patients go for care.
2. Proportional Flow Model: This divides patients among hospitals based on how many they admitted in the past.

Both methods assume that future patient behavior will not change. More complex models, like gravity models, use additional data but may be harder to understand.

In cases where admission data is not available, simple rules can be used. One approach assigns areas to the nearest hospital, while others may use a fixed distance, such as 40 km. More advanced methods include clustering techniques.

#Applications of Catchment Areas

Catchment area estimates are essential for predicting changes in hospital admissions. They can be applied in various settings, such as predicting emergency admissions in London or estimating malaria cases in Uganda. Despite their importance, few studies have looked into how the definition of catchment areas affects forecasting accuracy.

#Study Focus

In our study, we examined how different definitions of hospital catchment areas impact the accuracy of hospital admission forecasts during the COVID-19 pandemic, specifically from September 2020 to April 2021. We used a model that combined local COVID-19 cases with a delay from when cases are reported to when admissions occur. We aimed to see how varying definitions of catchment areas influenced the forecasts' accuracy.

#Data Sources

#COVID-19 Hospital Admissions

A COVID-19 hospital patient is anyone admitted after testing positive in the previous 14 days. Data on hospital activity, including admissions, is published weekly by NHS England.

#COVID-19 Cases

A confirmed COVID-19 case is an individual with a positive test result. Data on cases is compiled by local authorities and reported daily on the UK Government dashboard.

#Defining Hospital Catchment Areas

We used six different definitions to define catchment areas for NHS Trusts in England, focusing on where people go for hospital care. Each definition considered local authority boundaries and assigned a weight indicating the expected number of patients from that area to a given Trust.

The six definitions include:

1. Marginal Distribution: Uses historical data on hospital admissions to assign weights to local authorities.
2. Nearest Hospital: Patients go to the nearest Trust within their local authority.
3. Nearby Hospitals: Patients go to any Trust within a 40 km radius.
4. Emergency Admissions (2019): Based on historical emergency admissions data.
5. Elective Admissions (2019): Based on historical elective admissions data.
6. COVID-19 Admissions: Based on the distribution of COVID-19 admissions from local authorities during a specific time period.

#Comparing Catchment Area Definitions

We analyzed how the different definitions affected the forecasts. We looked at various statistics, comparing how many local authorities were included in each definition and how similar the areas were. The aim was to understand how differences in these definitions impacted forecasting results.

#Statistical Similarity

To measure how similar two definitions were, we used an overlap-similarity metric. This metric helps determine how much the areas defined by one method match those defined by another. The results indicated that some definitions were closely related, while others were very different.

#COVID-19 Case Trends

We also examined how trends in COVID-19 cases varied across local authorities. We expected that catchment areas with similar case trends would produce more accurate forecasts. In general, we found that local authorities with similar trends reported cases that were often strongly correlated.

#Forecasting Hospital Admissions

We used the convolution model to forecast hospital admissions based on the identified COVID-19 case numbers and the established delay in patient admissions. This model had previously shown good performance in making accurate forecasts.

We looked at two scenarios:

1. Retrospective Scenario: We used actual future COVID-19 cases for the most accurate forecasting.
2. Real-Time Scenario: We used estimates of future cases, which introduced uncertainty into the forecasts.

#Assessing Forecast Performance

We evaluated how well each catchment area definition performed in terms of accuracy. We calculated how often the forecasts contained the true values they aimed to predict. We measured how well the model accounted for uncertainty in the predictions.

#Calibration of Forecasts

Calibration measures how accurately a model predicts uncertainty. We calculated empirical coverage, comparing the proportion of forecasts that contained the true values. A well-calibrated model should match the predicted confidence intervals with actual outcomes.

#Probabilistic Forecast Accuracy

To evaluate probabilistic forecasts, we used the weighted interval score (WIS). This score considers how wide the forecast intervals are and whether the actual values fall within those intervals. It helps us understand the accuracy of forecasts over time.

#Results of the Study

#Differences in Forecast Accuracy

Our findings revealed notable differences in how the various definitions of catchment areas performed in making forecasts. The marginal distribution method generally led to the least accurate forecasts, while definitions based on COVID-19 admissions data provided the most dependable predictions.

#Correlations Among Forecasts

Despite variances between definitions, forecasts derived from different definitions tended to be closely correlated. However, specific dates saw significant deviations, especially when the COVID-19 case trends varied greatly across regions.

#Implications of Findings

The effectiveness of forecasting models is closely tied to how catchment areas are defined. Using a definition that considers local context, especially for specific health situations like COVID-19, significantly improves forecasting accuracy.

#Future Research Directions

Our study opens pathways for further research. It suggests examining other diseases or health crises to determine when catchment area definitions matter most. We can design simulations to test how varying case correlations within catchment areas impact forecasting success.

Additionally, we should explore whether simple heuristics can serve effectively in different contexts, especially when detailed data is not available. Understanding these dynamics will help health authorities use resources wisely in responding to health emergencies.

#Conclusion

In summary, defining hospital catchment areas effectively influences the accuracy of forecasts for hospital admissions. Our analysis during the COVID-19 pandemic showcased the importance of using context-specific data to enhance hospital planning and resource allocation. By understanding local patient behavior and trends, we can significantly improve health outcomes and preparedness for future healthcare challenges.