This meeting was organised by Professor Rumana Omar (UCL) on behalf of the RSS Medical Section. It was planned to be held at Errol St, but was unfortunately changed to a meetings venue at Barbican due to building works at the YMCA next to RSS.
Professor Nick Freemantle (UCL) talked about work which he and his co-authors had published as an ‘Analysis paper’ in the BMJ in 2015  as an update to a previous original research article . The earlier paper had not generated much attention, but despite the 2015 update not even being published as original research, it was used by the Secretary of State for Health as evidence to fuel an existing plan to rearrange the NHS system, including contracts for ‘junior doctors’, rendering weekend working much more part of the ‘normal’ working week. This became central to the arguments in the 2016 ‘junior doctors’ strike’.
The team analysed Hospital Episode Statistics data on all hospital admissions in 2013/4 in England relating the day of week (DoW) of admission to the risk of death within 30 days – whether in or out of hospital. The model took account of many confounding variables expected to affect this outcome, including demographic data, the urgency of each admission, whether each patient had had previous complex or emergency admissions, other factors describing ‘case mix’, and the DoW of each death compared with current DoW of all those still surviving (a ‘time-dependent covariate’). The fully fitted model took 12 days to run before delivering the results, with a surprisingly low 6.6% of cases with missing data. Although far fewer admissions and fewer deaths occurred at the weekend (Sat/Sun), there was a 10/15% increased risk of mortality within 30 days of admission for those who were admitted on a Saturday/Sunday, with a smaller increase on Mondays and Fridays, compared with Wednesdays (these effects all being highly statistically significant). From the same analysis, being in hospital on weekend days was associated with a slightly lower risk of death, if anything.
The speaker described several alternative or sub-analyses conducted to cross-check the results. The estimated 11,000 extra deaths associated with admission between Friday and Monday were used, often erroneously, as a political football. This was despite the care taken by Freemantle’s team to warn that although their results raised questions about “service provision at weekends”, further work was required in specific areas (for example the recent acute stroke care reorganisation) to examine whether deaths were avoidable. The authors were careful not to claim a causal link. In the way that this work was used politically, it was NOT a good example of the RSS aims of data and evidence leading to appropriate decisions!
Professor Freemantle described the costly personal consequences of the political sequelae of this research, including responding to 10 FOI requests, some about correspondence within his team. With hindsight he emphasised the value, when one is ‘in the centre of a media storm’, of excellent co-authors, and of meticulous QC of results prior to release.
Professor Richard Lilford (University of Warwick) spoke next about research that his team was currently doing, which had its roots in the ‘weekend effect’ analyses of Freemantle’s and other research groups. They had collated meta-analyses of similar research, and results concurred an increased mortality risk of around 11% for admissions at weekends compared with mid-week. He reiterated that there was no proven causal effect of weekend admission on 30-day mortality, and suggested that a low proportion of the estimated excess deaths may be preventable. Since the government had now implemented ‘7-day services’, including an increased presence of consultants in hospitals at weekends, his team was taking the opportunity to evaluate the effects and cost-utility of this aspect of the complex system of hospital care across all hospital trusts in the country over three years .
A proposed causal model of ‘difference in difference in differences’ was being used, incorporating changes in consultant presence, diagnostic errors, adverse events, patient length of hospital stay, death rates and quality of life. Costs of the changes in consultant activity and of patient social care and length of stay in hospital would be measured and any connections sought and tested. Expected biases of the model would also be quantified. Elicitation of estimates of key aspects of this model were being used to inform prior distributions for the model . A Bayesian elicitation exercise was conducted with nine ‘experts’ (familiar with clinical care at weekends and on weekdays) who judged that, on average, around 15% of excess weekend deaths in hospital were preventable. This prior belief will be updated as data for the model becomes available.
1. Freemantle et al. BMJ 2015; 351:h4596
2. Freemantle et al. J R Soc Med 2012; 105:74-84
3. Aldridge, Bion, et al. Lancet 2016; 388: 178-86
4. Lilford & Braunholtz. BMJ 1996; 313: 603
Report by Dr Gill M Price (University of East Anglia).