Research highlights – 2 September 2011

Research questions“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.

Diagnostic accuracy of screening tools for depression
This week’s contribution to the BMJ’s Research Methods and Reporting section looks at studies of diagnostic accuracy. These studies aim to compare the results of a diagnostic test or model with results of a reference standard in the same patients, yielding measures that include predictive values, post-test probabilities, ROC (receiver operating characteristics) curves, sensitivity, specificity, likelihood ratios, and odds ratios. Joris de Groot and colleagues use real examples to show how such studies can sometimes lead to biased and exaggerated estimates of diagnostic accuracy and, in turn, to “inefficiencies in diagnostic testing in practice, unnecessary costs, and physicians making incorrect treatment decisions.” One cause of bias is failure to adjust analyses to allow for the inclusion in such studies of patients who already have or are very likely to have, the disease.

Brett D Thombs and colleagues take this point further in their meta-review, focusing on diagnostic tools used to screen for depression. They scrutinised 17 eligible systematic reviews and meta-analyses, covering 197 primary publications, and found none commenting on possible spectrum bias from inclusion of patients who already had depression or were being treated for it. They are concerned that only eight of the primary studies specifically excluded such patients—it’s worth noting, though, that de Groot and colleagues counsel against such exclusion and argue instead for corrected analyses to account for the bias. Given these findings and the authors’ comment that “no clinical trial has found better depression outcomes for screened versus unscreened patients when the same treatment and care resources are potentially available to both groups,” should we be more worried about the ubiquity of depression screening in so many healthcare settings.

Health inequalities with cystic fibrosis
Survival among people with cystic fibrosis has improved dramatically over the past 50 years, with the median age at death in the UK rising from 6 months in 1959 to 27 years in 2008. Two long established factors associated with early death are low socioeconomic status and female sex. Helen Barr and colleagues used death registration data in England and Wales from 1959 to 2008 to investigate whether these socioeconomic and sexual inequalities in mortality from cystic fibrosis have weakened with the overall improvements in prognosis. Disappointingly, they found no substantial narrowing in the inequalities over the 50 year study period.

The cause of the sex gap in survival remains unclear, although David Taylor-Robinson and Michael Schechter suggest in their linked editorial that socially determined gender roles are probably as much to blame as biologically determined sex characteristics. The socioeconomic gradient in survival may be easier to understand.

Predicting death from cancer: man versus tool
How long have I got left? It’s not an unreasonable question for a patient with terminal cancer to ask. And it’s important too, because patients, relatives, and doctors need to make plans. But how comfortably, or—more importantly—accurately, do doctors answer ? Apparently, not well, and usually optimistically.

In an editorial Paul Glare writes that doctors tend to dodge questions of prognosis. Perhaps we can be forgiven; he goes on to say that we are rarely trained to do it, and there are no tools to predict death in widespread use. Those that do exist have methodological limitations, or have not been validated.

Gwilliam and colleagues used death markers identified in previous studies to create a tool that could predict the time to death as “days,” “weeks,” or “months,” more accurately than professional opinion. Their population was just over 1000 patients who were new to 18 palliative care services in England. The tool can function with and without blood results, and it is accurate about 60% of the time. But as the authors and Glare write, there are cautions—for example, it will need to be validated, and it is not very user friendly, although an app is in the pipeline.

Chocolate consumption and cardiometabolic disorders
A meta-analysis by Adriana Buitrago-Lopez and colleagues suggests that increased intake of chocolate is associated with a substantial reduction in the risk of cardiometabolic disorders. The association was significant for any cardiovascular disease (37% reduction), diabetes (31%), and stroke (29%), but not for heart failure. However, the authors found no randomised trials in their systematic review—only limited observational studies in selected populations—so much more evidence is needed to confirm whether eating chocolate is good for you.