Casemix adjustment

PROMs scores are adjusted using statistical models which account for the fact that each provider organisation deals with patients of differing complexity or Casemix. Some hospitals may treat older, sicker patients and others may be specialist centres that can take patients with more complex conditions than can be treated at a general hospital.

Casemix adjustment allows for fair comparison between providers and England as a whole. In effect it estimates the score an organisation would have if it treated a population with the same complexity as the national average.

The models were initially developed by the Department of Health and subsequently revised by NHS England. Currently hip and knee replacements are on the third generation Casemix adjustment methodology having been updated to account for the differences between primary procedures and revision procedures. 

Not all submitted data can be used. For a record to be successfully Casemix adjusted and to be included in the modelled output, several criteria must be met:

• the episode must link to a pre-operative PROMs questionnaire (Q1)
• the episode/Q1 linked record must link to a post-operative questionnaire (Q2)
• the episode/Q1/Q2 linked record must contain a complete measure (for example Oxford Hip Score) pre and post operatively plus all the necessary fields (e.g. IMD) required of the Casemix model.

Due to data quality issues and because participation in PROMs is voluntary, modelled records are typically around 40 per cent of all eligible episodes, although this varies by organisation and by procedure type. The following diagram illustrates this.

Casemix adjustment calculation is a two-stage process. Firstly, individual records have an ‘expected’ post-operative score created and secondly these expected scores are used to create an ‘adjusted’ score, aggregated at organisation level. Adjusted scores are only calculated where there are at least 30 modelled records. The statistical models break down with fewer records and aggregate calculations on small numbers may return unrepresentative results.

The expected score is a value that the models predict that a patient with certain characteristics will score themselves post-operatively based on the national distribution. These characteristics are based on variables from the patient’s HES episode and from the information that the patient has completed on the PROMs questionnaire.

Only items that have been shown to have a significant effect on the patient’s outcomes are used and these are different for each of the PROMs procedure types and measure. Expected values are based on a starting constant value and are modified, either positively or negatively by each individual variable until the final expected value is calculated. Variables can either be continuous, such as age or pre-operative score in which case a multiplier is applied, or discrete such as a diagnosis code of arthritis (yes/no) whereby a constant is applied.

For example, if it has been found that on average female patients score higher than males for a certain condition and measure, this would have the effect of increasing the expected score of female patients relative to male patients. Likewise, if on average, asthma patients score lower nationally than non-asthma sufferers, this would have the effect of lowering the expected score for a patient who has indicated on their questionnaire that they suffer from asthma.

Aggregation is the process by which a final adjusted post-operative score is created for each organisation to allow fair comparison. The calculation looks at the average difference between the expected values and the actual values at record level and adds this to the national average

Adjusted Q2_{(Trust X)} = Average Q2_(England) + Average (Actual Q2_{(Trust X)} – Expected Q2_{(Trust X)})

The adjusted health gain (the difference between pre and post-operative scores) is simply calculated as the difference between the Adjusted Q2 and the England average Q1. 

Adjusted Health Gain_{(Trust X)} = Adjusted Q2_{(Trust X)} – Average Q1_(England)

If, on average, patients for a given organisation score themselves higher than their expected scores, that organisation’s adjusted score will rise above the national average and conversely, if the average score is lower than the average expected score, the organisation’s average will fall below the national average.

Patient outcomes will vary for a wide range of reasons, outside of the control of the provider trust. This random variation in patients means that small differences in averages, even when casemix adjusted, may not be statistically significant. Control limits are calculated to identify providers that are significantly better or worse than England as a whole. These are known as outliers.