General Practice - Practice Team Information (PTI)

PTI statistical notes

Standardisation

The occurrence of many health conditions varies by factors such as age, gender and/or the relative level of deprivation of populations. The PTI sample population is broadly representative of Scotland in each of these factors. Nonetheless, to adjust for small differences in age, gender and deprivation (as measured by the quintiles of the Scottish Index of Multiple Deprivation (SIMD) last updated in 2006) between patients registered with practices in the PTI sample and the total number of registered patients for Scottish practices as a whole, a method of weighted extrapolation equivalent to direct standardisation is applied to produce all the estimates published here. Normally standardisation would use a 'standard' population (as defined in epidemiological literature), whereas in this case the 'standard' population is the Scottish population of all patients registered with a Scottish general practice. This 'Scottish population' does not have a constant, pre-defined, even distribution over age and gender groups. Rates are calculated within each age-gender-deprivation category of the PTI sample and these rates are then applied to the number of people within the corresponding category of the Scottish reference population. The resulting estimated numbers are then aggregated to the desired level. For example, to show figures by gender, the subcategory estimates would be summed over all age and deprivation categories within each gender. To then show rates, the estimated (gender-specific) numbers are divided by the total number of people in Scotland within each gender.

The aim of the standardisation is to make the estimates reflect more closely the Scottish population make-up, rather than the sample population make-up, and is not aiming to put sub-categories on equal footing. For example, there is (naturally) a large difference between males and females in population size over 75, so even if the actual estimated number of patients with a condition in both gender categories is the same, the rates per 1,000 registered patients will likely be very different. Therefore, if comparing the all-ages rates for a condition particularly affecting the elderly, the differences between men and women are likely partly due to differences in age.

Confidence intervals

Most of the figures shown on the PTI web pages are estimates of unknown values and should not be confused with the (unobserved) 'true' value itself. Some estimates are more reliable than others due to factors such as sample size, completeness of data and consistency in data recording methods. A common way to indicate the precision of an estimate is to use confidence intervals. A confidence interval gives a range of values in which an estimate lies, along with the probability that the exact value will lie within that range. In general, the higher the probability and smaller the interval, the more accurate the estimate will be. In this publication most estimates are shown with a 95% confidence interval, i.e. there is a 95% chance that the 'true' value will be in between the lower and upper limits shown in brackets after the estimate.

To be able to calculate a confidence interval, it is essential to have a measure of the variation (the variance in statistical terminology) between the measurements that feed into the calculation of the estimate. In case of the PTI estimates, these measurements are taken from the individual practices, so if we for example calculate the rate of patients seen for diabetes in age group A, gender B and deprivation category C, we have 47 (the number of practices in 2007/08) measurements to calculate the overall value if all these practices have registered patients in that particular age/gender/deprivation category. If this is not the case, rather than assuming the missing measurement equals zero, we assume the rate would equal the estimate based on the other measurements. We can then calculate the variance of these measurements assuming the practice observations are normally distributed.

Because practices vary considerably in size, measurements from each practice are weighted by the size of their population. Therefore it is not possible for an individual practice to dominate the results more than can be justified by the relative size of its population. If estimates are shown at an aggregated level, for example over all age or deprivation categories, standardisation is carried out within practices. This means that a contribution is calculated from that individual practice to the overall value, standardised by (in this case) age and deprivation, which is then weighted by the practice's size. The variance is calculated over the standardised practice contributions, weighted by practice size. This process is explained in more detail in the document "Updating PTI estimates" which is available from the PTI team on request. This document also shows the differences between this new calculation method and the old method with regard to overall number of patients seen for a number of common and less common conditions. The old method did not make any adjustment for practice size and effectively assumed that the patient population was sampled independently from practices. Although the changes could sometimes be substantial, the 'old' estimate was outwith the newly calculated confidence interval only for a single condition (dementia). This may be due to this condition being very strongly associated with age in combination with some practices having very atypical age distributions.

Main contact: PTI queries

 Printer friendly version