The quality of information on monitoring for haematological adverse drug reactions

Abstract

Aims

To examine the adequacy of instructions to monitor for haematological adverse drug reactions in the Summary of Product Characteristics.

Methods

We searched the United Kingdom eMedicines Compendium to identify instructions to monitor for haematological adverse drug reactions, and selected 84 Summaries of Product Characteristics for nonhaematological drugs, which were then scored independently by five clinicians, using a scale we devised, the Systematic Instructions for Monitoring (SIM) score. A subset of comparable summaries from Australian and United States summaries was also examined.

Results

The SIM scores for the five clinicians agreed well: Kendall's coefficient of concordance = 0.937, P < 0.0001. The median SIM score for the 84 UK summaries was 13 [95% confidence interval (CI) for median 12, 15] out of a possible 31. Over 40% fell below a hypothetical minimally acceptable score of 12/31. SIM scores were on average 2.0 (95% CI 0.4, 3.8) higher for Australian summaries; US summaries had intermediate scores that did not significantly differ from those in Australia or the UK.

Conclusions

Instructions on monitoring for adverse drug reactions are often inadequate. Pharmaceutical companies and regulatory agencies should produce clear guidance on monitoring, and the data to support it.

Introduction

Monitoring is a process of checking a system that is already active to ensure that the system is effective or safe. Monitoring for adverse effects by repeated laboratory testing can, under appropriate circumstances, anticipate or detect early signs of adverse drug reactions and prevent serious outcomes [1]. The earliest practical laboratory testing seems to have been designed to test whether a patient prescribed the antibacterial drug chloramphenicol could be developing bone-marrow toxicity of the type known to occur at high dosage and to be reversible; another type, which could occur at any therapeutic dosage, generally resulted in fatal aplastic anaemia in spite of monitoring [2].

We considered the information necessary for satisfactory monitoring according to a simple algorithm of the process (Figure 1). The clinician responsible for monitoring a particular laboratory variable, such as white cell count, needs to know the purpose of monitoring if it is to be meaningful. He or she also requires an indication of the acceptable values of the variable in a patient receiving treatment. Some reactions occur at the onset of treatment and others (such as heparin-associated thrombocytopenia) only after several doses of treatment have been given. Reactions can also cease when treatment ceases, or become manifest or continue for some time after treatment ceases [3]. The clinician therefore needs to know when to start monitoring, how frequently to test, and when it is safe to cease monitoring. Having obtained a result, the clinician needs to be able to interpret it and to base an action on the interpretation. For example, the action may be to continue treatment if the value is acceptable, otherwise cease treatment, but instructions to act will often be more complex.

Figure 1.

A simple algorithm for monitoring drug treatment. It assumes that monitoring for one specific variable starts before treatment and continues at predefined intervals until treatment ends; and that the acceptable range of values for the variable is known

The Summary of Product Characteristics (SPC, also known as the datasheet or drug label) is a regulatory document that may contain advice on monitoring. We examined whether SPCs from the UK gave sufficient guidance to allow satisfactory monitoring, as judged by the extent to which the required steps were explained. For comparison, we also examined a number of similar documents from Australia and the USA. We confined our study to monitoring for haematological reactions for non-oncological drugs in the general patient population, excluding, for example, monitoring instructions specific to patients with renal failure.

Methods

We constructed a list of the information necessary to allow satisfactory monitoring, and from this, a Systematic Information for Monitoring (SIM) score, composed of seven domains: why to monitor, what to monitor, when to start monitoring, how frequently to monitor, what to look for, how to respond (each scoring between 0 and 5) and when to stop monitoring (scoring 0 or 1). We used the unweighted aggregate SIM score (maximum 31) as a measure of the extent to which the instructions might allow satisfactory monitoring. Experienced prescribers could infer some of the aspects of monitoring – e.g. that monitoring should start when treatment started – so we also considered whether the instructions provided a minimum of evidence for an experienced prescriber. We considered that the SPC should give information on what to monitor, how frequently to monitor it, what to look for, and how to respond (four essential domains). Each domain would have to be at least reasonably specific (e.g. ‘monitor full blood count’, rather than ‘monitor blood tests’), giving a score of at least 3 per domain. We therefore took SIM scores below 12/31 (= 4 × 3) to be unacceptably low.

We identified relevant UK SPCs from an electronic database [4]. Each of five scorers then independently calculated a score for each relevant SPC. We used Kendall's coefficient of concordance to test the stability of the scoring method. We compared a subset of UK SPCs with the corresponding Australian and US documents. Not all drugs or SPCs were available in all countries, so we used a Generalized Linear Model and Tukey's method for multiple comparisons to analyse the international data (Minitab® Release 14; Minitab Inc., State College, PA, USA).

Results

We identified 1450 UK SPCs whose section on monitoring (Section 4.4 of the SPC) contained one or more of the terms count*, monitor*, blood, platelet*, or clotting. Of these, 84 relevant SPCs fulfilled the entry criteria and were scored by all five scorers. There was good agreement on SIM score between the five scorers (Kendall's coefficient of concordance = 0.937, P < 0.0001). Individual scores are shown in Figure 2. The median SIM score was 13 [95% confidence interval (CI) for median 12, 15, distribution interquartile range 6–21]. SPCs with low summary scores contained phrases such as ‘periodic blood cell counts… are called for’. The SIM score was not correlated with the date on which an SPC was authorized (r = 0.048, P = 0.49) or revised (r = 0.26, P = 0.26). Three scorers graded US and Australian documents for drugs nationally available from a subset of 13 drugs. Scores differed significantly among countries (P = 0.017). SIM scores were on average 2.0 (95% CI 0.4, 3.8) lower for UK documents than for Australian ones; US document scores were intermediate.

Figure 2.

Individual Systematic Information for Monitoring (SIM) scores of Summary of Product Characteristics (SPC) data on monitoring for each of five observers vs. mean SIM score of 84 United Kingdom SPCs (lower panel). Calculated SIM score for a subset of documents from the UK, USA and Australia, each representing up to 13 drugs and three observers (|) and 95% confidence intervals for differences from UK mean (◂▸) (upper panel)

Discussion

Optimal monitoring schemes would be reliable – detecting and preventing all potentially severe reactions; convenient – being neither unpleasant nor labourious for those involved; and cost-effective – repaying their cost in health benefits. A monitoring scheme cannot be regarded as being satisfactory without clear operational instructions. We have defined a series of steps that are required for satisfactory monitoring, and developed the SIM score as a simple quantitative measure of the extent to which a given set of instructions provides details of the required steps. All the steps are required for satisfactory monitoring. On the face of it, the score is therefore a valid index of whether the instructions are satisfactory. The score also showed good interrater reliability for the five different observers who assessed the instructions. More sophisticated scoring systems, giving different weights to different steps in the process, would be possible.

We found that SPCs differed greatly in the extent to which they gave adequate instructions on monitoring. An SPC containing outlines essential for monitoring should score at least 12, yet 37/84 (44%) scored less. We found a sample of UK and US documents similar and Australian ones slightly better. The statistically significant difference of SIM score for the Australian SPCs, which were on average 2 points higher than the UK ones, may be clinically insignificant.

Well-specified instructions are necessary but not sufficient to ensure that monitoring schemes will work. Effective schemes require, in addition, a satisfactory frequency of monitoring, accurate tests, established ranges of acceptable values, robust action plans, and careful implementation. Monitoring can work: the haematological monitoring of treatment with clozapine (Clozaril®), average SIM score 29, prevents 90% of cases of fatal agranulocytosis [5].

We found that SPC descriptions of monitoring schemes are often too vague to be useful. Pharmaceutical companies and regulatory agencies should produce clear guidance on monitoring, and the data to support it.

Competing interests

M.P. is a member of the Committee on Safety of Medicines (CSM), and M.P. and R.E.F. are members of the CSM Subcommittee on Pharmacovigilance, but the views expressed here are their own. M.P. has current grant support from Astra-Zeneca, Pfizer, and Bristol Myers Squibb for research into adverse drug reactions. J.C., S.A.C. and A.R. have no competing interests.