Imperial Research Proposal Example: Health applicant deciding MPH or health data route (Score 93)

Health systems in high-income countries increasingly generate administrative data across large populations, yet the translation of that data into actionable policy remains uneven. In England, NHS secondary uses data — including Hospital Episode Statistics (HES) and linked primary care records — have expanded substantially since the Health and Social Care Act 2012, creating a richer evidence base for population-level analysis. Despite this expansion, it remains unclear whether policies that mandate integrated data reporting at the integrated care system (ICS) level have produced measurable reductions in preventable hospitalisation, a widely used proxy for primary care effectiveness and care coordination. This proposal asks: To what extent has the introduction of integrated care system data-sharing frameworks in England, between 2019 and 2023, been associated with changes in ambulatory care sensitive condition (ACSC) admission rates, and do these associations vary by deprivation quintile? The question is bounded and testable. ACSC admissions are defined by a well-established NHS classification and are routinely reported at sub-national level. The 2019–2023 window captures both the pre-ICS baseline and the first operational years of the ICS model, providing a natural quasi-experimental structure. The deprivation dimension is not decorative: if data integration benefits accrue disproportionately to less deprived areas — because wealthier ICSs have stronger data infrastructure — then the policy may widen rather than close health inequalities. That is a question with direct implications for NHS England's Core20PLUS5 programme. Two bodies of literature bear on this question. The first concerns the effectiveness of integrated care models. Work drawing on the King's Fund evaluations of integrated care pilots, and on comparative studies of accountable care organisations in the United States, suggests that structural integration alone does not reliably reduce hospital utilisation; the mediating mechanism appears to be information sharing and care coordination capacity. Kodner and Spreeuwenberg's foundational framework distinguishes organisational, functional, and clinical integration, and subsequent empirical work — including studies of the Canterbury model in New Zealand — has found that functional integration, particularly data linkage, is the dimension most consistently associated with reduced emergency admissions. The second literature concerns data-driven quality improvement in the NHS. Researchers working with HES data have demonstrated that ACSC admission rates are sensitive to primary care access and to the quality of chronic disease management. Studies using difference-in-differences designs around earlier NHS reforms — including the Quality and Outcomes Framework introduction in 2004 — have shown that administrative data can identify policy-attributable changes in admission patterns when the reform has a clear implementation date and geographic variation in uptake. The gap between these two literatures is specific: there is no published study, to my knowledge, that uses the ICS transition as a quasi-experimental event to test whether data-sharing mandates — as distinct from organisational restructuring — are associated with ACSC rate changes, stratified by deprivation. Evaluations of ICS performance published by NHS England and the Health Foundation have been largely descriptive, and the few quantitative studies focus on process metrics rather than admission outcomes. This proposal addresses that gap directly. The study will use a difference-in-differences (DiD) design with ICS-level panel data. The unit of analysis is the ICS (n = 42 in England), observed quarterly from 2017 Q1 to 2023 Q4. The pre-period (2017–2019) establishes baseline trends; the post-period (2020–2023) captures ICS implementation, with a sensitivity analysis excluding 2020–2021 to account for COVID-19 disruption to elective and emergency pathways. The primary outcome is the age-sex standardised ACSC admission rate per 100,000 population, drawn from publicly available NHS England statistical releases and, where access is granted, from HES aggregate extracts. Secondary outcomes include emergency readmission rates within 30 days and GP referral-to-admission conversion rates. The key independent variable is a composite data-integration index constructed from NHS Digital's annual Data Security and Protection Toolkit returns and ICS self-reported data-sharing agreement status. This index will be treated as a continuous treatment variable rather than a binary switch, reflecting the reality that ICS data integration has proceeded at different speeds across regions. Control variables will include ICS-level deprivation score (Index of Multiple Deprivation 2019 quintile), total registered GP list size, proportion of population aged 65 and over, and NHS provider financial position. Fixed effects for ICS and quarter will absorb time-invariant unobservables and common temporal shocks. Standard errors will be clustered at the ICS level. The deprivation interaction will be modelled by including a deprivation quintile × post-period interaction term, allowing the DiD estimate to vary across the socioeconomic gradient. If the interaction coefficient is negative and significant for higher deprivation quintiles, that would be consistent with the hypothesis that data integration benefits are inequitably distributed. All primary data sources are publicly available or accessible via NHS England's standard statistical release programme. Where HES aggregate data are required beyond published tables, an application to NHS England's Data Access Request Service will be submitted; this is a standard academic route and does not require individual-level data, substantially reducing the ethical complexity. The design is feasible within a one-year taught MSc research project. The publicly available NHS England statistical releases cover the full outcome and most control variables at ICS level; no primary data collection is required. The data-integration index construction is the most labour-intensive component and will require approximately six weeks of systematic extraction from NHS Digital's Toolkit portal, which publishes annual organisation-level returns. The principal ethical risk is misattribution: a DiD design can identify association but not mechanism, and the proposal is careful not to claim that data-sharing causes reduced admissions. The write-up will foreground this limitation explicitly and will not be used to advocate for specific NHS procurement decisions without further causal evidence. If HES aggregate access is delayed or denied, the analysis can proceed entirely on published statistical releases, with a reduced set of secondary outcomes. This is a genuine contingency, not a fallback that undermines the core question. Provisional timeline: months 1–2, literature review and data assembly; months 3–4, index construction and data cleaning; months 5–7, primary analysis and robustness checks; months 8–9, write-up and revision. This is consistent with a September start and a June dissertation submission. No individual-level patient data will be used. The study falls outside the scope of NHS Research Ethics Committee review under Health Research Authority guidance, but confirmation will be sought from Imperial's Research Governance team before data collection begins. Imperial's MSc Health Policy sits within the School of Public Health, which houses research groups working on NHS performance evaluation, health inequalities, and applied health economics. The programme's quantitative methods training — including modules in health econometrics and applied statistics — directly supports the DiD methodology proposed here. Access to the College's research computing environment will be sufficient for the panel data analysis, which involves a dataset of modest size (42 units × 28 quarters). The proposal draws on the kind of applied quantitative work that characterises the School's NHS-facing research portfolio, including work on ICS performance and primary care access. The research question is tractable within the MSc dissertation format, requires no fieldwork or primary data collection, and produces a defensible, policy-relevant output: an estimate of whether data-integration policy has reduced preventable admissions, and for whom. The expected contribution is modest and honest. A well-executed DiD study on 42 ICSs over seven years will not resolve the causal question definitively, but it will provide the first systematic quantitative evidence on the ACSC–data integration relationship at ICS level, stratified by deprivation — evidence that is currently absent from the published literature and directly relevant to NHS England's ongoing ICS evaluation programme.