“Research highlights” is a weekly round-up of research papers appearing in the print BMJ. We start off with this week’s research questions, before providing more detail on some individual research papers and accompanying articles.

• What factors are associated with diabetic ketoacidosis at diagnosis of type 1 diabetes in young people?
• How effective is real time continuous glucose monitoring compared with intermittent self monitoring in type 1 diabetes?
• Is vertebroplasty more effective than placebo for recent or painful osteoporotic vertebral fracture?
• Is there an association between pandemic influenza A (H1N1) 2009 vaccine and Guillain-Barré syndrome?
• What is the predictive power of existing, and recently proposed, stroke risk stratification schemes for observed vascular event rates in older patients with atrial fibrillation?

Ketoacidosis at diagnosis of type 1 diabetes
Why do so many children with type 1 diabetes present only when they’ve developed ketoacidosis? Studies have suggested a wide range of risk factors relating to clinical and personal characteristics and health services, but nobody to date has pulled all this evidence together.

Now Juliet Usher-Smith and colleagues’ systematic review has found that younger age (particularly <2 years old), misdiagnosis, delayed diagnosis, minority ethnicity, and lack of health insurance (at least in the United States) were consistently associated with ketoacidosis at the time of diagnosis of diabetes. But the quality of the primary research wasn’t great overall, hence the authors rightly describe their review as exploratory and suggest that further research focuses on developing age specific preventive interventions. The findings should be reasonably generalisable, however, as the review included 46 cohort studies from 31 developed countries covering more than 24, 000 children.

Meanwhile, might this paper also inform health education? The risk of ketoacidosis at diagnosis was lower for patients in circumstances where diabetes was more likely to be spotted—when they had better educated parents, had a first degree relative with type 1 diabetes, or were in a population with relatively high background incidence. Editorialist Sasigarn Bowden picks up the theme of raising awareness and cites an Italian campaign where education among doctors and the public was associated with the incidence of ketoacidosis in newly diagnosed type 1 diabetes plummeting from 78% to almost zero at the University Hospital of Parma (http://bit.ly/qaq0b7).

Glucose monitoring in type 1 diabetes
Earlier this year we published two crossover trials of the artificial pancreas, a closed loop system linking a continuous glucose monitor and a subcutaneous insulin infusion pump (BMJ 2011; 342:d1855). The widespread use of such a device is probably still a long way off, but what about the continuous glucose monitor alone, a technology that has been around for at least a decade?

In a BMJ paper this week John Pickup and colleagues note that “uptake of continuous glucose monitoring in clinical practice has been limited to date because the evidence for its effectiveness has appeared only recently and has varied between studies. As a result, funding from national health services and insurance reimbursement has been restricted.” In response to this uncertainty these authors conducted a meta-analysis of individual data from nearly 900 patients in six randomised controlled trials.

Their regression model found that, for a patient using continuous monitoring daily, HbA1c would fall by about 0.9% (9 mmol/mol) when the baseline HbA1c is 10% (86 mmol/mol) over a couple of months.  Median exposure to hypoglycaemia seemed to be reduced by about a fifth during continuous glucose monitoring compared with self monitoring of blood glucose. The included trials were not, however, set up or powered to study the incidence of hypoglycaemia.

The overall conclusions from this study are probably not definitive enough to change practice, but the data will now allow the cost effectiveness of continuous glucose monitoring to be calculated for different patient groups according to their baseline HbA1c percentage, usage of continuous monitoring, and age. Although this is quite specialist research, we thought its relevance to clinical researchers, health economists, and policy makers made it a suitable paper for the BMJ.

Interpreting borderline statistical significance
In the Research Methods and Reporting section, Allan Hackshaw and Amy Kirkwood explain why borderline significance in the primary end point of a trial does not necessarily mean that the intervention was ineffective. If you’re confused about P values and confidence intervals, read this and be reassured by the message that “the true effect of an intervention is more likely to lie around the middle of a confidence interval (that is, the point estimate) than at either end.”