Babylon GP at hand’s Antibiotic Prescribing Statistics Are Probably Misleading
Babylon GP at hand have published graphs appearing to illustrate their prescribing of antibiotics as being low compared to the national general practice rates. The data they present is quite accurate, but quite probably misleading.
The implication is that Babylon GP at hand are prescribing appropriately and efficiently for their patients. However, this data does not support that conclusion.
Their antibiotic prescribing rates are generated using something called STAR-PUs for antibiotics. The STAR-PU model was most recently generated in 2013 to allow comparison of general practice prescribing rates for antibiotics, factoring in the different demographics in each practice. In a nutshell, a practice population is divided into segments based on their age and gender and each segment gets a different weighting, this is then added together to produce the STAR-PU population (denominator). The total amount of money each practice spends on antimicrobial products (numerator) is then divided by the STAR-PU population.
STAR-PUs are better than comparing raw practice prescribing rates, but fail to factor in a whole raft of demographic data points (features) that would make the numbers more amenable to comparison.
In 2015, NHS England released a comprehensive model that identified the features that correlate with various health outcomes, including prescribing spend. They then used these correlations to produce a “need index” for each practice — a ratio that corrects a practice population with a greater degree of accuracy than ASTRO-PU (the STAR-PU for all drugs, rather than one specific group).
The features in this newer model include deprivation, fertility rates, long term condition rates, the very elderly and ethnicity.
Until very recently, Babylon GP at hand specifically excluded patients who are pregnant or have certain long term conditions. Additionally, older people make up a tiny proportion of their population.
Therefore, under this model, Babylon GP at hand’s practice population would be weighed more negatively. This would cause a reduction in the denominator in their prescribing rate, likely resulting in their “performance” appearing worse than it currently does.
I have, for illustration, applied the NHS England prescribing need formula to Babylon GP at hand using real numbers where available and estimates where not (methods described below).
However, there is a large caveat to be applied here. The NHS England model and the ASTRO-PU/STAR-PU models are all based on “typical” English General Practices. Babylon GP at hand is very much not a “typical” practice, and thus would be classified as an outlier and likely excluded when generating the models. As such, subsequently applying the models to Babylon GP would be inappropriate and would lead to misleading results.
Which takes us back to where we started. The STAR-PU antimicrobial statistics Babylon GP at hand are promoting are probably misleading as they are based on the assumption that it is a “typical” practice.
It is possible that newer models may give a better representation of their prescribing performance, however, due to Babylon GP at hand’s outlier status, even more comprehensive models risk producing misleadings results.
Methods
The following variables were inputted into the model spreadsheet produced by NHS England for Prescribing Need Allocations.
ASTRO-PU data was calculated using current age/sex population bands as provided by NHS England.
Proportion of over 75s who are over 85 — calculated using actual data.
Proportion of registered patients aged >70 years claiming Disability Living Allowance — CCG Average applied.
Standardised mortality ratio (all ages) — CCG Average applied.
Fertility rate: Number of births (2012–2014) / Number of females aged 15–44 registered with practice — Assumed to be low given pregnant patients advised to avoid the service, assigned nominal rate of 0.01.
Practices with largest proportion of registered 20–24 year olds (top 1%) — Assigned value of 1 on basis that population distribution very likely to fall into this criteria.
IMD overall score —CCG Average applied.
Age-sex standardised proportion with activity limiting health conditions — Assumed to be low given patients with long term conditions advised to avoid the service, estimated at half CCG average.
Proportion in social housing — CCG Average applied.
Age-sex standardised proportion non-white — CCG Average applied.
