“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.

• Are epidural steroid injections effective for patients with chronic lumbar radiculopathy?
• What is the risk of admission to hospital for hyperkalaemia in older patients treated with trimethoprim-sulfamethoxazole and spironolactone?  
• How does mammography screening affect surgical treatment for breast cancer?
• Does the inclusion of multiple data for the same outcome in trial reports affect the results of meta-analyses?

Should we stop steroid injections for radicular back pain?
A third of back pain is predominantly neuropathic, and steroid injections are an increasingly popular management strategy. A group of patients presenting with chronic lumbar radiculopathy entered a trial that compared sham injection with caudal epidural saline or steroid injection, conducted by Trond Iverson and colleagues. Powered at 80% to detect a 10 point difference on the Oswestry disability index, they found no clinically significant difference in outcomes.

Is this the end for steroid injections? It doesn’t seem so. These results add to a heap of other trials and meta-analyses with contradictory findings. Editorialist Steven Cohen has raked through the trial in detail, and for those who lack the time or expertise to do the same, he raises interesting questions. Did the investigators inject the best place, with the right volume and dose? To what extent did the population drive the results? It seems unlikely that this trial will dramatically alter his practice—what will it do for yours?

Surgery rates after breast cancer screening
Among the coauthors of this paper by Pål Suhrke and colleagues, the names of Jørgensen and Gøtzsche should be familiar to BMJ readers from several previous articles critical of mammographic breast screening programmes—in particular of the claimed benefits of screening. One of the supposed benefits of mammography screening is that discovering tumours at an earlier stage may reduce mastectomies by increasing the potential for breast conserving treatment.

However, this study of breast surgery rates during the stepwise introduction of screening in Norwegian counties finds an initial increase in mastectomies and an overall increase in surgery in the age group invited to screening (50-69 years). The authors suggest that over-diagnosis is the likely cause—but as Richard Smith discussed in his BMJ blog this week, little is known about the rate of natural progression of ductal carcinoma in situ (DCIS). Its incidence has rocketed, and when this is communicated to women in whom DCIS is discovered, many would rather have the lesion removed than live with uncertainty. A randomised controlled trial of watchful waiting and yearly mammography versus surgery for DCIS is under way, which may result in improved estimates of risk of natural progression of early precancerous breast lesions.

Multiplicity and meta-analyses
Meta-analyses are considered the highest form of medical evidence, but are general clinicians right to retain some scepticism? Extraction of trial data for meta-analysis can pose a problem, because trials often report the same data in several different ways—reflecting different intervention groups, time points, and outcome measures. Given a choice of possible numbers to incorporate, reviewers might select particular results according to unconscious (or conscious) preference, so there’s a risk of bias. Could this affect the conclusions of meta-analyses?

Britta Tendel and colleagues looked at data reporting for trials included in a random sample of Cochrane reviews. They found that presentation of results for multiple subgroups, time points, and scales was common, and that trial protocols often lacked information about which data to choose. They also showed that the results of meta-analyses could vary substantially depending on which data were included.

So it does seem that bias in data selection may skew what meta-analyses say—a finding that might seem obvious, but is not well documented. The authors comment, however, that their random selection methods might not reflect how data are chosen in real life; for example, some specialties might have unwritten rules about which outcomes to pick. They suggest that to reduce potential bias, authors of systematic reviews and trial protocols should provide explicit strategies for handling multiplicity of data.

Views and experiences of men who have sex with men on the ban on blood donation
As the government in England, Scotland, and Wales announces that the ban on blood donation by men who have sex with men will be reduced from lifetime to a year, Pippa Grenfell and colleagues present qualitative evidence supporting the move.

Time trends in mortality in patients with type 1 diabetes
Valma Harjutsalo and colleagues report that survival of people with early onset type 1 diabetes has improved in the Finnish population, owing to a decrease in chronic complications, whereas survival of those with late onset type 1 diabetes has deteriorated since the 1980s, with an increase in deaths caused by acute diabetic complications.

Mortality after hospital discharge for people with schizophrenia or bipolar disorder
There is a persistent and increasing gap in mortality between discharged psychiatric patients and the general population, conclude Uy Hoang and colleagues in this retrospective study of English data.

Dedicated outreach service for hard to reach patients with tuberculosis in London 
Mark Jit and colleagues assessed the cost effectiveness of the “Find and Treat” service, which aims to identify and manage patients with active tuberculosis in populations with social risk factors, such as homeless people and those with drug problems.