Alex Nowbar reviews the latest research from the top medical journals
The biostatistician backlash—researchers’ inappropriate requests for data analysis
Have you been mis-sold study data by researchers? Even if you critically analysed it, you were likely mis-sold it. You could be entitled to your money back if your claim involves falsifying statistical significance to support a desired result. If you think you were mis-sold study data, act now and we’ll interpret your claim on the basis of expectation, not actual results. Start your claim now you can find out for free if the power calculation was really a priori or post hoc. Don’t miss the claim deadline or we’ll ignore key missing data that could bias the results.
Wang et al’s survey of 400 statisticians provides worrying insight into the mis-selling scandal that is going on in research today and is a must read. The statisticians weren’t asked what they did with the inappropriate requests, but a glance at the state of research reporting might help you extrapolate.
A confounder conundrum—safety of beta blockers in pregnancy
It is a truth universally acknowledged that pregnancy makes treatment of any medical condition anxiety-inducing for patient and clinician alike. Isn’t it lucky then, that this paper is about beta-blockers? I jest—beta-blockers aren’t how I would start to treat anxiety and the reason for treatment in this large cohort study was actually hypertension.
So to the point, Bateman et al focus on beta-blocker use in the first trimester in pregnant women with a diagnosis of hypertension in Denmark, Finland, Iceland, Norway, Sweden, and US. They assessed the risk of congenital malformations and concluded that there was no “large increase in the risk for overall malformations or cardiac malformations, independent of measured confounders.”
There will always be a long list of confounders for this type of study design. For example, they note that judging beta blocker exposure by filled prescriptions doesn’t mean the patient actually took the medication. So they re-ran the analysis where they defined exposure as at least two dispensings of the medication as refilling the prescription suggests the patient is taking the medication and they found similar levels of risk.
They also re-ran the analysis taking into account the potential for pregnancies ending in stillbirth or abortion that their original analyses wouldn’t have captured and again, found similar levels of risk.
Finally, they only considered women with hypertension to resist the confounding by the indication for needing an anti-hypertensive. However confounding by a hypertensive patient’s blood pressure being high enough to require treatment means there will be unmeasured differences between the exposed and unexposed groups that could influence the results. Still, in practice, decisions must be made and this is the best available evidence at the moment.
The Roux-en-Y reassessment—which measure of weight regain is useful
Look to this study, not for how it will affect you or your practice (it won’t unless you are a patient or researcher of bariatric surgery), but to admire the authors’ concern for which outcome of weight regain after bariatric surgery is associated with outcomes that are important for patients. King et al conclude that their findings “may inform standardizing the measurement of weight regain in studies of bariatric surgery”—a laudable aim given how many different ways of measuring weight regain there are.
There’s weight, BMI, weight regain as a percentage of presurgery weight, weight regain as a percentage of nadir weight and thresholds of these such as greater than 10kg, greater than 5 BMI points, 10% of presurgery weight and so on. They compared 13 different measures of weight regain for association with progression of diabetes, hyperlipidaemia, and hypertension and declines in physical and mental health-related quality of life and satisfaction with surgery. Weight regain quantified as percentage of maximum weight lost came out top. This may disappoint some readers as it requires more than one step of calculation.
The machine manages—CT head analysis by deep learning algorithms
The time is ripe for deep learning for diagnosis. Chilamkurthy et al designed algorithms for nine “critical findings” on the extremely ubiquitous test, non-contrast CT head. They base the “gold standard” on reporting by three independent radiologists. The authors validated their algorithm on scans from two different datasets—one at another centre, but the algorithm did not perform as well there. The authors proposed two slightly tenuous reasons for the poorer performance—first that the radiologists of the other centre’s dataset had access to the clinical history and second that the majority decision of three radiologists used for their primary validation dataset is a “better gold standard.” This begs a few questions. Would the algorithm have performed better if it learned from more radiologists? Can a gold standard truly be defined without seeing what happened to the patients? The authors conclude that the algorithms could be used for triage which presumably means a radiologist would still report the scan. Their algorithms probably meet the bar for this application.
It is also worth mentioning that the study was funded by a company called qure.ai that employs almost all the authors, but I suppose that’s who does this work. I suspect there will be several more bridges to cross before this can be implemented in clinical practice. A lot of this will be about trust and credibility. Which a Lancet paper can bring. But independent validation and expert scrutiny would help, and ideally before the inevitable randomised controlled trials, to avoid exposing patients to unwarranted risk. Will there be a regulatory body for this soon?
An unambiguous undoing—clindamycin for bacterial vaginosis in pregnancy
The French government-funded PREMEVA trial is a breath of fresh air. This large double-blind randomised controlled trial found no evidence of reduced risk of late miscarriage or spontaneous very preterm birth with screening for and subsequent treatment of bacterial vaginosis. The treatment was clindamycin and those who received it had significantly more diarrhoea and abdominal pain than those randomised to placebo. Cancelling out an indication for antibiotics—great. Filtering this into clinical practice—priceless.
JAMA Internal Medicine
The hip-hop high—tobacco and marijuana products in music videos
There are two types of journal readers—homies, who think “I must read this” when they see hip-hop music videos in the title, and not-at-homies, who don’t. After reading the title, I anticipated something quite light-hearted. But Knutzen et al found that 40-50% of videos (analysed from 2013 to 2017) contained combustible or electronic tobacco or marijuana use with the more popular songs (by number of views) having a higher prevalence of use. Apparently, this corresponds to over 39 billion views. No one can quantify the impact this level of visibility has on public health but it must be enormous. There was a particularly dramatic rise in the appearance of branded electronic products from 2013 to 2017. The discussion was interesting. The authors discuss the introduction of warning statements mandated by the US Food and Drug Administration (FDA). However, more ethical music producers may want to ban this type of product placement altogether.
A daring decision—attitudes of over 65s to deprescribing
Many clinicians observe some patients who resist deprescribing and this may be discouraging them from deprescribing for other patients to avoid conflict. That is why this paper on the attitudes of older Americans is so important. It shows that the majority are willing to have medication deprescribed if their physician said it was possible and that more than two thirds want to reduce the number of medicines they are taking. I sincerely hope this research encourages open discussion between clinicians and their patients to reach a decision that is patient-centred.
Immunotherapy for advanced breast cancer
In this double-blind randomised controlled trial, the pharmaceutical company “collaborated” with the academic steering committee on the trial design, data collection, analysis, and interpretation. Trials are hard to run without sponsorship, but if conflicts of interest were DEFCON levels this trial would probably be at 1 i.e. can’t get more conflicted than this.
No surprise then that the trial is painted positively and referred to in the press release as “a massive step forward” when all it showed was an improvement in progression free survival (hazard ratio 0.8 with p=0.002) but not overall survival (hazard ratio 0.84 but p=0.08) and more adverse events with the new treatment (atezolizumab). The press release says “we are able to significantly extend lives.” Whose interpretation was this? Progression free survival is when the disease on the scan doesn’t worsen and the patient doesn’t die. The press release comment seems to refer to actual survival which was not statistically significant in this trial. The survival did look better in the subgroup positive for the receptor that the drug is supposed to bind to (PD-L1 positive).
Presenting these results as more than they are can give false hope to patients, clinicians, and researchers.
The other one—cyclin dependent kinase inhibitor for Advanced Breast Cancer
This double-blind randomised controlled trial has a different drug, different breast cancer type, and a different pharmaceutical company, but again the sponsor was a part of the steering committee and was involved in critical aspects including collection and interpretation of data. Again there was no statistically significant difference in overall survival, but a positive interpretation of the subgroup who were sensitive to previous endocrine therapy that did have better survival. What happened to progression-free survival you might ask? Well, that was presented in a previous paper and was positive, but this alone has no meaning for patients. Again the press release declares that this treatment should be the standard of care.
Pre-eclampsia and vascular dementia
Pre-eclampsia is a fascinating, but also very serious condition manifesting as hypertension and organ dysfunction in pregnancy. Basit et al conducted a large prospective cohort study with long follow-up to look at the association of pre-eclampsia with later dementia in Denmark. They found women with a history of pre-eclampsia had more than three times the risk of vascular dementia. This is a lot higher than I would have guessed even though vascular dysfunction is well known to underlie pre-eclampsia. And the association persisted despite adjustment for comorbidities such as cardiovascular disease, diabetes, and hypertension.”
These data are interesting and should inform research into pathways, but the authors’ conclusion as to the applications of this data in clinical practice may slightly overstep the mark. History taking for pre-eclampsia is unlikely to be of use at the moment because we don’t screen for vascular dementia. We don’t screen for it because, as per Wilson-Jungner criteria for validity of screening, even if we identify the condition early, we don’t have much in the way of disease-modifying treatment.
Alex Nowbar is a clinical research fellow at Imperial College London.
Competing interests: None declared