Monitoring oral anticoagulant therapy: Measuring coagulant activity

Jørn Attermann

The aim of this Ph.D. thesis was to examine the reliability of the International Normalized Ratio (INR) which is recommended by the World Health Organization for monitoring patients on life-long oral anticoagulant therapy (OAT) with vitamin K antagonists. This therapy is offered to patients with increased risk of thrombosis, e.g. patients with artificial heart valves or atrial fibrillation, but close monitoring of the INR is necessary to avoid life-threatening bleedings.

Traditionally, patients on OAT are required to pay regular visits to a physician who decides on drug dosage adjustments and the time for the next visit based on laboratory analyses of the INR. This conventional treatment regimen is, however, relatively inconvenient for the patient, since it requires frequent outpatient visits and venipunctures. Moreover, errors may occur due to insufficient communication between patient and physician. There is general agreement that the quality of the therapy is too low, and often unexpected fluctuations in the coagulant activity are seen. In addition, numerous reports on inconsistent or discrepant INR values suggest that the distributional properties of the INR system are not fully understood.

Recently, by the availability of small portable whole blood coagulometers, OAT based on patient self-management has become a realistic alternative to conventional therapy. This new concept is more convenient for the patient and may also increase the treatment quality.

The thesis provides a basis for evaluating and comparing INR values obtained by traditional plasma-based methods and by the new generation of whole blood coagulometers. The specific hypotheses were:

• The precision of patients’ own measurements of INR performed at home on a portable coagulometer is sufficient to allow for self management of OAT (first substudy).
• For selected pairs of thromboplastins, the relation between logarithmic prothrombin times is linear, such that the standardization by the INR system is feasible (second substudy).
• Based on a realistic statistical model, the statistical distribution of INR estimates, given a true INR value, can be adequately described to allow for good estimates of accuracy and precision (third substudy).

In the first substudy it was shown that for selected patients the precision of the patients' own measurements of INR is sufficient to allow for reliable routine patient self testing of INR. In the same substudy we found large discrepancies between the INR measurements on portable coagulometers and in the Department of Clinical Chemistry, Skejby Sygehus. However, when available estimates of laboratory and portable coagulometer variation were taken into consideration, the observed differences were not unlikely. In the same study we found that the deviation between the portable coagulometers and the Department of Clinical Chemistry, Roskilde County Hospital, was not larger than the deviation between the two clinical chemistry departments. The material in our study is, however, not strong enough to draw any firm conclusions regarding the accuracy of neither portable coagulometers nor clinical chemistry departments.

In the second substudy we investigated the fundamental assumptions of the INR system. We found that data from the comparison of three thromboplastin preparations (CRM 149S, Nycotest and Hepato Quick) were consistent with these assumptions and concluded that the INR system may be valid for these thromboplastins.

In the third substudy we proposed a statistical model for INR estimates obtained in a randomly chosen laboratory, given the true INR is known. This model was based on our experiences in the second substudy, as well as on a number of reports on international calibration exercises. The model describes and quantifies central aspects of the INR system, such as the inaccuracy of INR estimates based on a given path of calibrations. The main result states that, under weak regularity conditions, log(log(estimated INR)) is approximately normally distributed with mean log(log(true INR)). The variance is a function of the interlaboratory variance, the accuracy of the slope estimates in the calibrating laboratories, the accuracy of local prothrombin time (PT) determinations, and the interindividual variance of the normal persons used for normalizing the PT. A simulation study showed that the derived theoretical results are in close agreement with the empirical ones. The third part of the thesis also contains a formula and test for the interlaboratory variation of slope estimates in a calibration exercise.

In an example, in which we investigated the first generation international reference preparation (IRP), RBT/79, we found that the 95% prediction interval for INR estimates, obtained directly by this IRP, is 1.7-2.4 and 3.2-5.0 for the true INR being 2 and 4, respectively. These intervals are large, compared with the therapeutic range of 2-3 for patients with mechanical heart valves. In routine use, the prediction interval is even larger, since routine INR estimates are obtained by commercial thromboplastins. The International Sensitivity Index (ISI) of these thromboplastins are assigned either by the use of plasma calibrators, which have had their INR determined against an IRP, or by successive calibrations against a house standard and an IRP. Moreover, the imprecision of IRPs will generally increase with the distance from the primary IRP 67/40.

Based on results from the statistical model it is recommended that in calibration exercises an effective standard error of the ISI is reported instead of the traditionally reported marginal standard error. The effective standard error contains sufficient information on the interlaboratory variance, such that it can be used directly for the calculation of the coefficient of variation of INR estimates. In the example given above, the effective standard error of the ISI is 0.0956, and in comparison the marginal standard error is only 0.0364. Thus, the use of the marginal standard error will result in gross overestimation of the precision of INR estimates. This may explain some of the discrepant and "inexplicable" INR values, which are frequently reported in the scientific literature.