On Wednesday 17 June, 70 attendees joined an event jointly hosted by Leeds-Bradford local group and the RSS Medical Section. The theme for the day was ‘Mortality statistics’ and had presentations from Gary Childs and Professor Mohammed A Mohammed.
Gary Childs from NHS Digital presented on the Summary Hospital-level Mortality Indicator. The SHMI compares the number of actual patients who die following hospitalisation to the number that would be expected to die given the patient’s characteristics. The SHMI was developed after 2011 following a steering group and an expert technical group who defined the requirements for a mortality statistic. Subsequently, NHS Digital were commissioned to operationalise the specification within the NHS.
The process for calculating the SHMI start with integrating HES data with ONS data, and applying some filters such as specialist trusts and stillbirths. The modelling approach then uses three years of data to inform 142 logistic regression models for a set of diagnosis group, eg TB and heart valve disorders. Many of the variables are consistent across diagnosis groups, such as sex, age, comorbidity, and admission method. Finally, the SHMI statistic is computed by dividing the number of observed deaths by the number of expected deaths and often plotted on a funnel plot with 95% control limits. Importantly, the SHMI is used an indicator to prompt further investigation (of data quality or hospital performance), rather than a measure of avoidable deaths or ranking statistic.
Gary finished his talk presenting recent improvements and future plans for SHMI. A recording of the talk is available on the Leeds-Bradford local group website and the local group’s playlist within the RSS YouTube channel.
Our second speaker was Professor Mohammed A Mohammed, who spoke about ‘(Mis)Understanding Hospital Mortality Statistics’. In particular, Mohammed spoke about the conceptual underpinnings of hospital mortality statistics. Questions were raised as to whether mortality statistics are interpreted as an indicator of avoidable deaths, despite them not being designed with that intention. An example from primary care as given that highlight misuse and misinterpretation of mortality statistics, partly explained in his 2004 publication. The Case-Mix Adjustment fallacy was presented also, which notes that outcomes are functions of at least patient case-mix, chance, and quality of care - the last of which can only be measured by proxy in statistical models. Thus, accounting for some of these variables does not always appropriately permit attribution of residual variance to quality of care.
Author details
Ciarán McInerney, PhD, is secretary of the Leeds-Bradford local group. He is a research fellow in the School of Computing at University of Leeds and the NIHR Yorkshire & Humber Patient Safety Translational Research Centre, where he studies the design and evaluation of digital innovation for patient safety.