You are here
Modelling
Distribution modelling and allocations.
The allocation of postgraduate medical training posts to NHS Local Offices is one of the key duties of Health Education England (HEE), now the NHS England Workforce, Training and Education Directorate (NHSE WTE). This includes the training of appropriate numbers of specialists aligned to local population health needs.
At the heart of the Distribution Programme is the principle of ensuring equal opportunity of access for equal need in the allocation of specialised training posts. This approach is informed by NHSE WTE’s mandate to reduce inequalities regarding patient access to health services and ensure high quality of patient care.
The revised allocation formulae are based on assessment of factors such as demography, morbidity, deprivation, and projected service demand. This produces a distributional guide or Fair Share model for each region.
The Distribution Programme considers the 65 General Medical Council (GMC) recognised specialties (and 31 sub-specialties). There are separate utilisation-based formulae for each specialty. Allocations of speciality posts will therefore vary as a result of speciality specific adjustments, such as in the case of highly-specialised services which may see out of area patients accessing care.
The focus of the Distribution Programme is only on postgraduate medical tariff-funded posts. The approach to allocations does not include Trust funded, National Institute for Health and Care Research (NIHR)[1] funded or Defence Deanery[2]funded posts in baselines.
Post allocations are predicated upon a weighted formula comprised of three utilisation-based components.
Component 1 – NHS Resource Allocations Methodology Core Allocations
Allocation needs are calculated on an Integrated Care Board (ICB)[1] footprint. To calculate the need of service provision for ICB areas the population for each age-sex group per GP practice is multiplied by the relative care need per patient as estimated from research. The products for each age-sex group are then aggregated to give the relative need-weighted population for each GP practice. The weighted populations for GP practices per footprint are totalled to give the relative need-weighted populations for the region.
Outlined below are three approaches for specialty allocations where post will move region as result of this Fair Share modelling. These are: General and Acute, Mental Health, and Maternity.
[1] https://www.england.nhs.uk/publication/integrated-care-boards-in-england/
General and Acute
Relative need per head for General and Acute services is estimated using a person-based approach, first developed by the Nuffield Trust[1]. The person-based approach uses anonymised data at the individual level to provide accurate estimates of need for small and atypical populations.
The General and Acute model covers inpatient spells (continuous stays) in hospital and community settings, outpatient attendances, accident and emergency attendances and Critical Care. Activity in Maternity, Mental Health and community settings are not included as they are covered by other component models. Specialised services are subject to different commissioning arrangements and so are also excluded from this component.
Need estimations within the General and Acute model relative need are estimated from historical patterns of health service utilisation. For each individual registered with a GP practice, per-capita costs are calculated by applying a cost to each inpatient spell, outpatient attendance, A&E attendance and critical care day. The costs referenced within the modelling are projected using National Tariff prices where available, and otherwise reference costs. In a small minority of cases, the specialty average was used in the absence of tariff prices and reference costs.
Statistical modelling is then used to select the ‘best fit’ drivers for relative costs at the level of the individual and the relative weights for each driver. The quantified relationships produced from this statistical model are taken to be predictors of relative future, cost-weighted need for health care services, with the exception of the supply variables[2].
The modelling tested a wide range of potential variables to select those that were the best in statistical terms, and also plausible indicators of need, to be included in the final model. Morbidity (previous diagnoses) and age were the most important variables in the model.
[1] https://www.nuffieldtrust.org.uk/sites/default/files/2017-01/person-based-resource-allocation-web-final.pdf
[2] Supply variables are included in the NHSE Core Allocations models to account for the greater availability of healthcare services generally leading to greater utilisation of these services. The effect of these variables is to ensure that parts of the country that have less healthcare capacity—typically rural, coastal, and island populations—are not unfairly penalised in these formulae. Examples of supply variables include travel duration to hospital sites, median waiting times, bed and operating theatre numbers.
Mental Health
The adult mental health component uses person-level data on the use of Mental Health services, Learning Disability services, Improving Access to Psychological Therapies (IAPT) services, and General and Acute hospital services, as well as demographic characteristics and area level socio-economic characteristics. Specialised Mental Health services, which are commissioned by NHS England, were excluded from the model whenever data would allow.
The model is based on national datasets that capture person-level service usage in a consistent and comparable way. The two main datasets used were the Mental Health Services Data Set (MHSDS12) and the IAPT dataset. Information on inpatient and outpatient care were complemented with Secondary Uses Survey (SUS) data when not reported in the MHSDS.
Individual cost-weighted activity estimates were calculated by aggregating the cost for inpatient bed days (split by the cost per general bed day and cost per intensive bed day) and unit costs for community care contacts (split by the pay band of the care professional overseeing the care) and IAPT contacts. These data were merged with other person and area level information derived from other routinely collected data available within NHS England for all individuals registered with a GP practice in England.
The Mental Health model is subdivided into two further components: an adult model and a Children and Young People (CYP) model. The adult model concentrates on those aged 20 and over, whilst the CYP model concerns those aged 19 and younger. The method used all mental health activity captured as inpatient bed days and outpatient appointments within the Mental Health Services Data Set (MHSDS).
Cost-weighted activity estimates were calculated by quinary age groups and across these age groups by gender. The unit costs used for bed days and appointments were taken as estimated for the adult model. 9% of all the cost-weighted activity within the MHSDS was for CYP and the remaining 91% captured services delivered to adults, however, it should be noted that this activity excludes adult IAPT contacts. The CYP cost-weighted activity estimates were then expressed as a percentage of cost-weighted activity for the 20-24 age groups, split by gender. The percentage of cost-weighted activity by gender and for the four CYP age groups were then applied to the need per head from the refreshed adult Mental Health model by gender for those aged 20-24. These ratios were validated against data extracts from Hospital Episode Statistics (HES).
Maternity
The Maternity model uses person-based data to estimate cost per birth by GP practice. This included diagnoses in previous years and a wide range of data including from the ONS Census of Population.
A number of new variables were created for the refresh of the maternity component, including the proportion of births that were low birth weight births and the number of previous births by the mother. A smaller set of variables were tested for inclusion in the model than for general and acute, based on the plausibility of relevance for maternity services. For example, the proportion of those aged over 65 in the small area claiming state benefits was not tested. Age and some morbidity markers (previous diagnoses) were found to be important determinants of predicted costs per birth. Supply variables were included in the model but set to the national average in the calculation of weighted populations.
ONS statistics on live births by age of mother were used to weight female registered patients aged 15 to 44 by quinary age bands within LSOAs, to distribute live births in England by GP practices. The costs per birth were applied to the average annual number of live births. For practices that had opened since the end of 2013/14 (and for which there is not a cost per birth available from the model), the average cost per birth for the LAD in which the GP practice is located was used.
Component 2 – Future Demand Projection
The approach taken in the Fair Share modelling to estimate need-weighted populations for regions is to multiply the population for each age-sex group by the relative need for inpatient and outpatient care per head estimated from historical patterns of health service use. This historical usage data is drawn from NHS Hospital Episodes Statistics (HES) data. The products for each age-sex group are totalled to give the relative need-weighted population of each ICB sub-region; the weighted populations for each ICB sub-region then are aggregated to give the relative need-weighted population per region.
Within the Fair Share modelling, where population growth in an area is disproportionate in a younger or older population – which will affect relative levels of need – this is reflected in the changes in need-weighted populations over time. The person-based usage of inpatient and outpatient care was assumed to not change over time.
Estimated population growth was generated using Office for National Statistics (ONS) projections for resident populations in ICB footprints by quinary age-sex group between 2018 and 2042. The ONS projected populations are the 2018 based Sub-National Population Projections (SNPPs) published at Clinical Commissioning Groups (CCGs) age-sex level[1] (CCGs were dissolved in July 2022 and their duties taken on by the new Integrated Care Boards (ICBs)).These projections start with the 2011 Census populations, which are rolled forward to 2018 by adding the number of births and net migration and subtracting the number of deaths. Trends for the fertility rates, death rates and net migration are used by the ONS to project forward from 2018. The equivalent rates using data obtained from the 2021 Census are unfortunately not yet available but will be taken into account in future modelling.
An adjustment is made in the Fair Share modelling allocations component to account for health inequalities and for unmet/under-met needs and health inequalities. This adjustment is based on the Standardised Mortality Ratio (SMR) for those under 75 years of age (SMR <75). The advantage of this measure is that it is relatively stable within each geography over time, and it is relatively straightforward to understand and interpret. SMR <75 does however also present some disadvantages: it relies on premature mortality being a reliable proxy for morbidity as well as mortality inequalities. This is not always the case, e.g. for mental health and musculoskeletal conditions where disease may result in disability rather than death. Furthermore, SMR <75 relates only indirectly to health inequalities experienced by persons over 75 years of age. As the programme proceeds, we will continue to evaluate whether to retain SMR <75 or to replace it with an alternative metric.
Component 3 – Adjustment for the Provision of Highly-Specialised Services
The third component of the Fair Share model concerns the number and proportion of inpatient and outpatient episodes funded by Specialised Services in each NHS Trust.
Data for this component is drawn from Trust-level activity data from NHS Hospital Episodes Statistics (HES).
The weighting of component three relative to the other two components is proportional to the percentage of inpatient and outpatient care in the given speciality commissioned by Specialised Services as opposed to by ICBs. The weighting of this component is higher in specialties more exposed to Specialised Services commissioned activity, such as those that diagnose and treat cancers, genetic disorders, and some complex medical and surgical conditions. The maximum weighting of this component has been capped at 25% of the overall model.