The aim of this article was to compare clinical versus medication - based identification of populations of patients with diabetes and hypertension and the effects each of them could have on the calculation of P4P indicators. We showed that there was a quite good degree of data concordance and that the results of the P4P indicators obtained with the two methods differ little in terms of final scores (difference of 4% in favour of the medication-based identification) but that these results correspond to different realities.
We observed that both of these two modes of identification can cause the target populations to be under-estimated, either – to a small extent – due to the doctor forgetting to code the morbidities, in the case of clinical identification, or – to a greater extent – due to the fact that patients whose only treatment is life style and diet rules are not taken into account in medication-based identification. Hence, in both cases the level of the indicator can be over-estimated. In our study, the rate of forgetting to code the morbidities in the clinical approach is 4.9% for diabetes and 8.2% for hypertension. These results are consistent with those already published in the literature assessing the completeness and correctness of computerized general practice medical records where the rate of forgetting to code the morbidities was situated between 5 and 10% 
. Medication-based identification under-estimates the population of diabetic patients by 27.8% because there are patients who are given life style and diet rules only. The percentage of such patients is higher in our sample than it is in the French literature, where it oscillates between 10% and 15% 
without any clear explanation for this. However it goes along with the results of a cross-sectional study of 253 618 patients lead in the UK in 2004 which showed that 31.3% of all patients with type 2 diabetes were being managed with diet only 
. Concerning patients with hypertension, the medication-based identification under-estimates this population by 15.8% due to patients following lifestyle and diet rules only, and to the well-known therapeutic inertia in hypertension. This rate of patients with untreated hypertension is lower than that reported in a large country-wide study on general practice in France where one third of the 70,000 studied had never received hypertension medication 
. This rate was also estimated at 27.5% in a US study 
The medication mode of identification can on the other hand result in an over-estimation of the target population of patients with hypertension, and consequently in the under-estimation of the level of the indicator, since medications for hypertension may be indicated to treat morbidities other than hypertension. In our study the identification of patients with hypertension on the basis of medication led to the inclusion of 12.8% of patients who did not have hypertension and should not have been included. To our knowledge this has never been underlined before. A perverse effect of this medication-based identification, in the framework of a P4P programme, could be to expose these patients to the risk of being prescribed statins by non scrupulous doctors for the sole purpose of improving their individual scores.
From an economic point of view, if we were in the framework of a P4P programme, the mode of identification would not substantially change doctors' financial remuneration. This is because the final scores obtained look “quite similar” when we compare the clinical and medication-based methods whereas they deviate from the target objectives by over 20%. However, even if we cannot make any statistical comparison between the results obtained with the two methods, when we look at the confidence intervals calculated for the two indicators selected, the medication-based identification method appears in our sample to be more advantageous for doctors. This is probably due to the fact that the diabetic population treated by lifestyle and diet rules only, which is huge in our sample, is not included in this method of calculating the indicator.
Limits of the study
This study has several limits. 1/ For the medication-based approach, the target populations were identified on the basis of medications prescribed by doctors, not on the basis of refunded medications as it would be in the case of a P4P programme. Unfortunately we did not have access to the Health Insurance reimbursement database. No study today enables us to assess the extent of the gap there may be between medications prescribed and medications bought by the patients and reimbursement by health insurance. We can posit that it is narrow in these populations of patients with chronic diseases who tend to be compliant with their doctor's prescription. 2/ We cannot conclude on the statistical significance (or not) of the difference between the scores according to the mode of identification, as the construction of the indicators was based on different populations. We can only conclude that the difference is of 5% in our sample, and that the confidence interval is not large, especially since the number observations (number of doctors, n
61) is not big. 3/ Data collection for this study was carried out on a volunteer basis and not in the framework of P4P. Judging by the British case, it is possible that clinical under-coding exists with P4P 
. This under-coding would tend to reduce the difference between the two types of identification by artificially improving the indicator calculated on the basis of clinical identification. 4/ The correlation between the two search modes in this sample of French practices is strikingly good and probably reflects good use of diagnostic computer entry in these patients who attend on a regular basis for routine monitoring. This may not be generalizable across other conditions.
Our findings do not enable us to conclude that one of the two identification methods is better than the other. The two approaches yield very similar scores but these scores cover different realities and offer food for thought on the possible usage of these indicators in the framework of P4P programmes. Although it may seem reasonable to use these indicators in order to compare a doctor's activity from one year to the next regardless of the identification mode, provided that it remains the same. On the other hand, using the absolute value of an indicator seems meaningless either for estimating the intrinsic quality of care delivery, or to compare doctors between each other because it depends, among other things, on how it was built.