Another month, another episode – episode #2, to be specific. Available for download or in iTunes.
HealthDay says: “Opioids No Better Than Ibuprofen for Pain After Car Crash: Study”, leading with an assertion that prescription painkillers are no more effective than non-steroidal anti-inflammatory drugs. This was also picked up by the daily American College of Emergency Physicians e-mail newsletter.
So – no?
Despite the best of intentions, there is simply no reliable conclusion to be drawn from the cited publication. In the citation, the authors perform a propensity score-matching secondary analysis of prospectively collected observational data on patients discharged from the Emergency Department following a motor vehicle collision. There were 948 patients in their initial study cohort, with approximately half receiving a prescription at ED discharge. Propensity score matching then further excluded approximately 100 more, and finally patients lost to follow-up reduce their ultimate sample to 284. Their primary outcome was the presence of persistent self-reported moderate to severe pain six weeks after their MVC.
Unsurprisingly, with the wide confidence intervals mandated by their small sample, there was some overlap between the number in each group having persistent pain at six weeks. Thus, this leads the authors to make a guarded, but clearly anti-opiate, conclusion the evidence does not exist to recommend opiate therapy at ED discharge.
The bias in any underpowered study is to commit Type II error, which, as a reminder, is to retain the false null hypothesis in failing to detect an effect. Furthermore, as the authors note in their extensive methods section, in non-randomized studies, the measured and unmeasured confounders ultimately guide group assignment, which can bias the downstream results. The adjustments of propensity matching attempt to control for these, but tend to depend on large sample sizes and robust feature sets to reduce the magnitude of systematic bias – neither of which are present here. The need to impute missing data further reduces the reliability of under foundational data. Lastly, is their primary outcome relevant and related to the interventions examined? I am doubtful that six week persistent pain accurately reflects the scope of benefit (or lack thereof) relating to analgesic pharmacotherapy following MVC.
Avoiding the adverse effect of opiates is certainly important. However, this article should add little to the discussion – despite its lay medical press coverage.
“Persistent pain after motor vehicle collision: comparative effectiveness of opioids versus non-steroidal anti-inflammatory drugs prescribed from the emergency department—a propensity matched analysis”
Treatment evidence regarding venous thromboembolism is not particularly sparse – except what to do about calf VTE. The most robust evidence is three decades old, and of little use for generalizing to modern diagnostic methods and direct oral anticoagulants.
This, then, is the CACTUS trial – a randomized, double-blind, placebo-controlled trial examining the need for treatment of isolated calf DVT with subcutaneous nadroparin. The primary outcome was a composite measure of extension of calf DVT to proximal veins, contralateral DVT, or symptomatic pulmonary embolism. Safety endpoints were bleeding, death, and treatment-related adverse events.
Sadly, this evidence is mostly bereft of guidance. Over the six-year course of this trial, they screened 746 patients and only enrolled 259 – 50% of their goal sample. There were four (3.3%) VTE in the nadroparin group compared with seven (5.4%) in the placebo cohort, and these differences failed to reach statistical significance. Furthermore, clinically significant bleeding occurred in one patient in the nadroparin group, along with one clinically significant adverse medication reaction (heparin-induced thrombocytopenia). Thus, the authors conclude: “Nadroparin was not superior to placebo in reducing the risk of proximal extension or venous thromboembolic events in low-risk outpatients with symptomatic calf DVT, but did increase the risk of bleeding.”
However, half the patients enrolled had deep muscular DVT, further reducing the risk profile of their already grossly underpowered cohort. Thus, the question remains open – and probably the most relevant evidence would come from an adequately powered trial comparing the natural course of disease to both oral antiplatelet agents and direct oral anticoagulants.
“Anticoagulant therapy for symptomatic calf deep vein thrombosis (CACTUS): a randomised, double-blind, placebo-controlled trial”
Thanks to Tom Deloughery (@bloodman) for his insights!
Emergency physicians work holidays. It’s just part of the job description. I’ll be in the Emergency Department tomorrow, myself.
Which holidays are typically the lightest – according to this retrospective review from a single pediatric Emergency Department?
- Probably no different: Washington’s Birthday, Memorial Day, Columbus Day, New Year’s Day
- Probably lighter: Martin Luther King Day, Labor Day
- Almost definitely lighter: Thanksgiving Day, Christmas Day
Hopefully we’ll all have good shifts together tomorrow!
“Predicting Flow in the Pediatric Emergency Department: Are Holidays Lighter?”
The natural world is replete with bacteria.
Humans have existed on this planet for millennia.
In the ages before antibiotics, many humans succumbed to bacterial infections – while, of course, the vast majority survived.
This is not a profoundly reliable observational study, but it does help reinforce this basic concept. This report is a secondary analysis of the GRACE-10 study, which involved primary care patients recruited with a diagnosis of acute cough. The original study was a randomized, placebo-controlled trial for non-specific lower respiratory tract infection, as part of a genomics analysis for evaluation of antibiotic resistance.
This analysis, however, looks solely at the placebo arm, and examines the symptom course and resolution of those who were ultimately diagnosed with a bacterial cause of their LRTI and compares the with those who were not. Of the 834 patients included in their analysis (those with complete symptom diaries), 162 were thought to have a bacteria pathogen based on respiratory culture, nasal swab, or whole blood antibody titers.
S pneumoniae and H influenzae were the most common bacterial pathogens, with most of the remainder the “atypicals” for community-acquired pneumonia. And, at the end of the day: virtually everyone did fine. Patients with a confirmed bacterial pathogen in the setting of their LRTI improved slightly more slowly than those without, had more re-visits in follow-up due to worsening or new symptoms, and a greater percentage were placed on antibiotics in follow-up (12% vs. 6%). The remainder eradicated their bacterial pathogens without antibiotics – you know, the way humans and other contemporary mammals survived for eons.
Now, some of these cases positive for LRTI may be colonization and not pathogenic infection, while some of the negative cases were not diagnosed due to lack of sensitivity. But, regardless, the overall point of this article is probably valid – some bacterial infections will worsen, but in the generally healthy population, a delayed-antibiotic strategy might be valid as an attempt to improve antibiotic stewardship.
“Disease Course of Lower Respiratory Tract Infection With a Bacterial Cause”
Coronary artery disease – one of many self-inflicted wounds of Western society – fuels some of the largest pharmaceutical and device blockbusters of our time. Statins, stents, and the entire organization of our health system around STEMI care are all linked to coronary disease.
This JAMA article and its breathless lay coverage focus on a clinical trial for evolocumab (Repatha), one of the new proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors. This trial, featuring evolocumab added to a statin versus a statin alone, evaluated this therapy using one of the most surrogate of surrogate markers: nominal change in percent coronary atheroma volume at 78 weeks.
As the press releases indicate, this trial was a massive success – the $14,000-per-dose PCSK9 inhibitor was positive for its primary endpoint. Patients taking just a statin continued to have excellent LDL levels and their coronary atheroma volume, as measured by intravascular ultrasound, was essentially unchanged. The evolocumab cohort, however, had even better LDL levels and … coronary atheroma volume was essentially unchanged. But, the difference between +0.05% and -0.95% is statistically significant, and therefore, the trial was a success.
There were, of course, in this trial with only 968 patients, no signals of clinically relevant benefit nor obvious reliable harm. Considering the fierce debate regarding whether statins are already overprescribed, despite being ubiquitously inexpensive, I do not see any reason to look forward to this $14,000 drug entering more widespread use.
“Effect of Evolocumab on Progression of Coronary Disease in Statin-Treated Patients: The GLAGOV Randomized Clinical Trial”
There are two ways to treat a Nursemaid’s elbow (radial head subluxation) – supination/flexion or hyperpronation. I’ve done both. I’m a fan of hyperpronation, so, therefore, I’m highlighting a study that agrees with my practice pattern.
This systematic review covers 7 studies enrolling 701 patients, comparing the success rate and perceived pain of each technique. Trials were, generally and for some obvious reasons, limited in terms of blinding and outcome assessment. Pooled failure rate with hyperpronation 9.1% was while failure with supination /flexion was 27.3%. In studies reporting pain scores, subjective external rating of child pain during procedure also favored the hyperpronation group.
This video demonstrates both techniques, although I’ve seen variations on hyperpronation both using extension and flexion.
“Effectiveness of reduction maneuvers in the treatment of nursemaid’s elbow: A systematic review and meta-analysis”
The vast majority of the important evidence regarding the use of endovascular therapy for stroke has substantial limitations. The critical studies, with the largest magnitude of benefit, used strict imaging criteria to limit interventions to large-vessel occlusions with only small-volume ischemic cores surrounded by large regions of surviving tissue. Further generalizing these data to the remaining stroke population represents a significant challenge.
This small study tries to describe the benefit of endovascular treatment in a population with larger ischemic core volumes, specifically those greater than 50 mL – and it’s useless. They have 56 patients in their retrospective case-control comparison, and are missing long-term follow-up data for 9. Outcomes, yes, are better for the endovascular therapy group – a handful of patients had low or minimal disability, while none of the control patients achieved an mRS 0-2. Safety outcomes, of course, are a total wash in a small sample such as this. This would have made for a great conference abstract, but it is hardly compelling or significant data.
The main notable feature of this study is mostly how it reflects the real-world deployment of this therapy, regardless of the guidelines and current evidence. Many centers have expanded the use of endovascular intervention for patients beyond the scope of the original trials. These are very, very weak data – and, even though I don’t disagree in principle with imaging-guided revascularization, the further away from established evidence we drift, the lower value the intervention becomes.
“Endovascular Treatment for Patients With Acute Stroke Who Have a Large Ischemic Core and Large Mismatch Imaging Profile”
A couple weeks ago I covered computerized diagnosis via symptom checkers, noting their imperfect accuracy – and grossly underperforming crowd-sourced physician knowledge. However, one area that continues to progress is the use of machine learning for outcomes prediction.
This paper describes advances in the use of “big data” for prediction of 30-day and 180-day readmissions for heart failure. The authors used an existing data set from the Telemonitoring to Improve Heart Failure Outcomes trial as substrate, and then applied several unsupervised statistical models to the data with varying inputs.
There were 236 variables available in the data set for use in prediction, weighted and cleaned to account for missing data. Compared with the C statistic from logistic regression as their baseline comparator, the winner was pretty clearly Random Forests. With a baseline 30-day readmission rate of 17.1% and 180-day readmission of 48.9%, the C statistic for the logistic regression model predicting 30-day readmission was 0.533 – basically no predictive skill. The Random Forest model, however, achieved a C statistic of 0.628 by training on the 180-day data set.
So, it’s reasonable to suggest there are complex and heterogenous data for which machine learning methods are superior to traditional models. These are, unfortunately, pretty terrible C statistics, and almost certainly of very limited use for informing clinical care. As with most decision-support algorithms, I would be curious also to see a comparison with a hypothetical C statistic for clinician gestalt. However, for some clinical problems with a wide variety of influential factors, these sorts of models will likely become increasingly prevalent.
“Analysis of Machine Learning Techniques for Heart Failure Readmissions”
If you read the lay medical news, that’s the question being posed, indirectly, to Emergency Physicians at large. Why, oh why, are you terrible at accurately diagnosing and hospitalizing patients for cellulitis:
“Approximately One-Third Of People Diagnosed With Cellulitis Do Not Actually Have It, Study Suggests.”(HealthDay)
And, if you believe the authors of the cited article, cellulitis misdiagnosis leads to up to 130,000 unnecessary hospitalizations and $515M in avoidable healthcare costs, let alone the costs of various secondary harms.
Unfortunately, this well-covered indictment of our care of cellulitis comes from a not-so-reliable study: a retrospective evaluation of 259 patients hospitalized over a 2 year period. Of these patient charts reviewed, the authors felt as though 79 (30.5%) were misdiagnosed with cellulitis. The authors then plugged these numbers into their meandering cost calculations for unnecessary care, resulting in the numbers above.
The actual conclusions, however, bear no resemblance to the circulating headline. It is not “one-third of people diagnosed with cellulitis” – it is one-third of a narrowly defined cohort of hospitalized patients with lower extremity cellulitis. The misdiagnosis rate is based only on the shadowy shapes discernible through chart review, with all its omissions and inaccuracies. I doubt the sort of complicated medical presentations common at a teaching facility and referral hospital are generalizable to the vast majority of the ~2.5 million annual ED visits for cellulitis, most of whom are probably more straightforward. Then, even their cost numbers are probably inflated by using the average cost of an “unnecessary” 4.3 day medical stay – $12,656.90 – generalized to their cohort.
In many instances, in the setting of diagnostic uncertainty, it is absolutely reasonable to err on the side of caution and treatment. I am not certain these authors’ interpretation of their narrow slice of the healthcare spectrum accurately reflects the larger reality. Amusingly, though, their article ends with the statement “Our study serves as a call to arms for improving the care of patients with suspected lower extremity cellulitis.” Their proposed solution? A Dermatology consult on every case of suspected lower extremity cellulitis!
“Costs and Consequences Associated With Misdiagnosed Lower Extremity Cellulitis”