Published: 03/06/2004, Volume II4, No. 5908 Page 27
Statistical process control provides a sophisticated way of measuring mortality rates, says Roger Taylor
The Sunday Times' Good Hospital Guide, published last month, showed standardised mortality rates for NHS hospital trusts in England and Scotland.
The death rates show wide variations in performance at hospital trusts across the country. The figures measure mortality rates among patients suffering from the illnesses which cause 80 per cent of deaths in hospital, taking into account issues such as the age and sex of patients as well as whether admission was emergency or elective.
However, these figures disguise as much as they reveal.
They aggregate data across long time periods and many different departments. Although they can give a comparative overview, there is much that they miss.
For example, the one-time problems with paediatric heart surgery at Bristol Royal Infirmary would not have shown up on this type of measure.
A problem in just one department is disguised in the overall data. Similarly, a hospital may have significant problems for a short period but, again, this is hidden in a long-term measure.
A more sophisticated approach is to track mortality rates over time using statistical process control charts.
These can track outcomes day by day and month by month, revealing whether for any period of time death rates in a particular trust or particular department of a trust were out of line with the rest of the country.
Dr Foster monitored the last five years of hospital data to identify variations in performance between trusts in England and plotted the results. In the charts (right), the red line shows the degree to which the mortality rate is diverging from national averages. If the mortality rate measured reaches twice the national average with 95 per cent statistical certainty, the red line passes through the blue line. This is an 'alert' and indicates a potential cause for concern.
The charts are all based on the hospital standardised mortality ratio as published in The Sunday Times guide but tracked over time. The first chart shows a typical pattern for a hospital with worse than average mortality ratio over three years.
The second chart shows a pattern for a hospital with better than average mortality. The interesting point to note is that both hospitals have triggered alerts during this period - even the one with below-average mortality for the overall period.
The number of alerts ranges from zero at 15 hospital trusts to more than 50 at eight trusts.
The final chart shows the distribution of alerts across English trusts in this analysis.
Around 15 per cent of trusts account for half of the alerts triggered, although factors such as data quality and major structural or process changes within individual organisations are cited as reasons for poor results.
The data shows most trusts in England experience periods of high levels of risk in terms of increased death rates from time to time, as well as periods when the risk of death is significantly better than the national average.
Roger Taylor is research director of healthcare information analysts Dr Foster (www. drfoster. co. uk)