TL;DR
This paper introduces a novel data-driven method using multiscale community detection to define hospital catchment areas, moving beyond traditional single-hospital assignments. This approach better reflects patient behavior in complex urban healthcare markets, enabling more effective collaboration between hospitals to serve communities.
What Problem Does This Paper Address?
Understanding which populations a hospital serves (its "catchment area") is crucial for healthcare planning, resource allocation, and accountability. However, existing methods for defining these areas often struggle in urban environments where patients may visit multiple hospitals. Traditional approaches like "First Past the Post" (assigning a patient to their most frequently visited hospital) can lead to fragmented accountability and fail to capture the reality of patient flow. This paper tackles the challenge of creating mutually exclusive and collectively exhaustive catchment areas that accurately represent patient behavior, especially in areas with "dilute healthcare markets" where patients are distributed across several providers.
Methods
The core of this research lies in applying Markov Stability Multiscale Community Detection to patient admission data. Here's a breakdown of their approach:
- Data Preparation: They used anonymized Hospital Episode Statistics (HES) for planned orthopaedic surgeries in England from 2011-2015.
- Market Concentration Analysis: They calculated the Herfindahl-Hirschman Index (HHI) for each Middle Super Output Area (MSOA) to quantify healthcare market concentration (high HHI = concentrated market, low HHI = dilute market).
- Network Construction: They created an MSOA-to-MSOA similarity network. The "edges" between MSOAs were weighted based on the cosine similarity of their patient presentation patterns to various hospitals. Essentially, if two MSOAs send similar proportions of their patients to the same hospitals, they are considered more similar.
- Multiscale Community Detection: This network was then analyzed using Markov Stability. This method simulates how information (or patient flow) spreads through the network over "Markov time." Stable "communities" (groups of MSOAs) emerge where internal connections are strong and external connections are weak. The "multiscale" aspect means they identify these communities at various levels of granularity, from many small communities to fewer large ones.
- Catchment Area Definition: The identified communities of MSOAs were then assigned to the hospital located within that geographic region. Crucially, unlike traditional methods, these catchment areas can encompass multiple hospitals.
- Evaluation: The performance of these new catchment areas was compared to the traditional "First Past the Post" method using metrics like "coverage" (the proportion of patients treated by hospitals within their assigned catchment).
Datasets
- Hospital Episode Statistics (HES): Anonymized data on planned orthopaedic surgical admissions for adult residents in England between April 1, 2011, and March 31, 2015. This dataset includes information on patient residence (MSOA) and the hospital trust they attended.
- Office for National Statistics (ONS) Data: Used to obtain the population-weighted centroids of Middle Super Output Areas (MSOAs).
Key Insights and Ideas
- Beyond Single-Provider Catchments: The most significant insight is that catchment areas don't have to be tied to a single hospital. Multiscale community detection naturally identifies groups of MSOAs that share similar patient flow patterns, allowing for the creation of multi-hospital catchment areas.
- Reflecting Real Patient Behavior: By basing catchment areas on actual patient presentation data, the method captures how people actually access care, rather than relying on arbitrary geographic boundaries or simplistic "most frequent provider" rules.
- Urban Markets are Different: The method is particularly valuable in urban areas with dilute healthcare markets, where traditional methods fail to provide clear accountability. The new approach allows for collaborative coverage of these complex markets.
- Hierarchical Structure of Healthcare: The multiscale nature of the community detection reveals a quasi-hierarchical structure in healthcare access, allowing for different levels of analysis and policy decisions (e.g., local collaboration vs. regional organization).
Key Results
- Improved Coverage: The proposed method consistently achieved higher "coverage" (proportion of patients treated within their assigned catchment) compared to the "First Past the Post" method.
- Nationally, coverage increased from 76% (FPTP) to a maximum of 93.4% with 11 catchment areas.
- In the rural case study, coverage rose from 79% (FPTP) to 96.8% with 3 catchment areas.
- In the urban case study, coverage jumped from 58.4% (FPTP) to 96.3% with just 2 catchment areas.
- Fewer, More Representative Catchments: The method identified stable partitions with varying numbers of catchment areas, ranging from 127 down to 11 nationally. Coarser partitions (fewer catchment areas) generally led to higher coverage, as they naturally incorporated more hospitals.
- Data-Driven, Not Geographically Biased: The community detection process was based solely on hospital utilization data, not pre-existing administrative boundaries or geographic proximity, yet the resulting catchment areas showed a consistent alignment with geographical and organizational patterns.
Limitations and Future Work
- Subjective Interpretation: The Markov stability method produces characteristics of optimal partitions, but the final selection of the "best" partition requires subjective interpretation.
- Policy Context Needed: The authors emphasize that the resulting catchment areas need to be appraised within a wider policy context, considering local organizational challenges and priorities.
- Future Work: The paper suggests that this methodology can be a powerful tool for understanding how collaborations between hospitals and community care should be structured, especially as the NHS seeks to formalize local accountability. The ability to generate partitions at different granularities also opens doors for various policy purposes.
Summary updated