All glucose meters must be subject to formal quality control measures

EDITOR—A meta-analysis of diabetes care in the community concluded that unstructured care is associated with poorer follow up, worse glycaemic control, and greater morbidity than is in-hospital care.1 The article emphasised that the transfer of responsibility for diabetes to general practice without adequate support is associated with adverse patient outcomes. Such support is essential in the monitoring of the quality of community based glucose assays, although this was not specifically referred to in the article.

In 1987,as a consequence of a patient's death, the Scottish Health Service issued a safety hazard notice relating to the incorrect use of glucose meters.2 Key messages in the notice include ensuring that staff using the meters are fully trained and that quality control procedures are adopted. In response to this our laboratory developed a quality control scheme to monitor the performance of meters and users across our two hospital sites. The scheme entails distributing, fortnightly, sterile horse blood supplemented with glucose, and it has been effective in identifying poor user technique and malfunctioning meters, thereby helping to assure the quality of bedside glucose assays.

Recognising that glucose meters in use in general practice surgeries were not subject to formal quality control measures, which potentially puts patients in the community at risk, we decided to pilot our scheme to a general practice that was routinely using six glucose meters. Shortly after initiating the pilot scheme we identified a malfunctioning meter, which was giving an average systematic positive bias of 15% (table), and we arranged for its replacement. Internal quality control performance, for which the aqueous material supplied by the manufacturer was used, was within acceptable ranges throughout this period.

Two important issues emerge from this work. Firstly, all glucose meters, irrespective of their location, must be subject to formal quality control measures. Secondly, the quality of results cannot be assured by sole reliance on analysis of the manufacturer's aqueous quality control material. All laboratories involved in the quality control of glucose meters should consider extending such schemes into the community.

Like was not compared with like

EDITOR—Griffin's meta-analysis of diabetes care in general practice is timely.1 However, I have concerns, similar to those raised in the accompanying commentary by Greenhalgh, about some of his conclusions.

Only five studies met the inclusion criteria for the meta-analysis, and all patients were receiving outpatient care at a hospital. Griffin recommends that prompted care should be provided to a selected group of patients with diabetes. This prompting would need to be delivered to all patients with diabetes in the practice since it would be difficult to organise it for only a small proportion of patients. A study in which I participated found that just over half of all people with diabetes are under care in general practice, just under one fifth are under hospital care, and about one third are under shared care.2

To be able to draw valid conclusions from a meta-analysis, like must be compared with like. Other factors, such as the practice population and the organisation of the practice (other than recall systems), may also be associated with the quality of care. Although the proportion of local practices did not explain the interstudy heterogeneity in the meta-analysis,1 the size of the practice was only reported in two trials. 32 Only one study reported whether the practices ran diabetic “miniclinics.”4 Furthermore, as mentioned by Greenhalgh, the population in the studies included in the meta-analysis are subject to selection bias and are not, therefore, representative of the general population.

Many potential confounding variables have not been taken into account in the meta-analysis. I conducted a literature search on the quality of care of patients with diabetes in primary care and identified over 37potential confounding factors. It was inappropriate to use meta-analytical techniques for such a varied group of practices.

I am now involved in a study to determine which factors are related to good care of patients with diabetes. This study has collected data from 169practices (639general practitioners) in three different health authorities that have conducted audits of patients with diabetes. Preliminary results show that there are wide variations in process and outcome measures. Regression analysis indicates that having a recall system is not related to the process or outcome of the care of patients with diabetes.

Basing treatment decisions on evidence from randomised controlled trials is becoming increasingly acceptable in general practice. A randomised controlled trial with randomisation at practice level is required to assess the effectiveness of structured care with a prompted recall system.

Author's reply

EDITOR—Barlow and colleagues highlight the core point in my paper: trial data over 16years show the failure of unstructured diabetes care and support the need for registration, recall, and regular review if primary care surveillance is to be effective.1 Barlow et al extend the need for support beyond access to specialist advice to a welcome consideration of laboratory support for quality assurance of near patient testing of blood glucose.

Khunti, like others before him, underlines both the major contribution that general practice makes to diabetes care 2 3 and the point that the crucial ingredients of successful diabetes care have not been fully ascertained. However, it would be disconcerting if, on the basis of a regression analysis of audit data, Khunti was to advocate dispensing with recall systems. Particularly when the previous use of similar regression methods has explained only a small proportion of the variance in outcome.4

Khunti is concerned about selection bias. This should have been minimised by the quality of randomisation procedures and the concealment of allocation in the trials included in the meta-analysis. More important, however, is generalisability, which frequently limits the conclusions of trials and meta-analyses. It is relevant in this case because of the highly selected groups of patients and practitioners involved, the short duration of follow up, and the limited set of outcome measures. Consequently, generalisability merited a whole paragraph of discussion. The evidence from the systematic review was not interpreted as justifying the wholesale transfer of responsibility for routine diabetes care to general practice. Patients with complex problems who are registered at practices with little experience in diabetes care may be routinely followed up most effectively by a system of structured or prompted care based in a hospital diabetes centre.

Khunti raises again my own and Greenhalgh's concerns about the limitations of meta-analysis in this context. Of course adding apples to oranges does not make “appleoranges” but rigorous scrutiny of available data using a range of explicit methodologies does allow us to draw some conclusions about the likely benefits of fruit.

A study such as the one Khunti proposes would be large (in part because of the loss of statistical power inherent in cluster randomised trials), prolonged, and in danger of irrelevance. A spectrum of observational data—ranging from audit to aetiological epidemiology or, dare I suggest, meta-analysis using individual patient data—might most feasibly inform quality improvement in the organisation of diabetes care.