When doing more harms patients

Two trials reported in the last few weeks, in two completely different infections, point at the same uncomfortable idea. Some of the things we do for our sickest patients, that the guidelines tell us to do, and the observational evidence seemed to support, may have been harming, or even killing them.

Both of these findings are preliminary and not published in full. Findings from the REMAP-CAP trial have been presented at the CCRUK conference and the SNAP trial (adjunctive clindamycin arm) at ESCMID. But the direction they point is important, because the lesson is bigger than either drug.

REMAP-CAP is a large, multi-centre randomised trial for influenza and other respiratory infections (I am the lead paediatric investigator for my hospital). In the relevant part of the study, patients are randomised to 5 days of the antiviral medication oseltamivir (Tamiflu), 10 days, or no antiviral. This part of the study was recently paused after an interim analysis, usually indicating either a safety issue or that the drug could not reach the pre-defined superiority threshold.

The results have now been presented for those aged 12 years and older in the ICU - the sickest patients. In this group, 90-day mortality was higher in those who received Oseltamivir; roughly 19% for those randomised to 5 days and 22% to 10 days, against 14% with no antiviral. The adjusted odds ratios were around 2.3 in both treatment arms. The study was analysed using Bayesian methods, with a posterior probability of harm (increased mortality) at over 96%.

The trial team have since released an “information for healthcare professionals”Q&A with extra details, such as subgroups who seemed particualry high risk; those with shock, bacterial superinfection, symptoms for >5d, and the immunocompromised. They have explicitly recommends the drug not be prescribed for critically ill patients aged 12 and older with influenza.

I would like to re-emphasise that this is not a subtle signal. Oseltamivir is recommended for severe influenza by essentially every major guideline, and the evidence underpinning that recommendation has always been observational, and questionable. The most influential piece was the PRIDE collaboration’s individual-patient meta-analysis (Muthuri and colleagues, Lancet Respiratory Medicine, 2014), which pooled some 29,000 patients hospitalised with H1N1pdm09 across 38 countries and reported a 19% reduction in mortality with neuraminidase inhibitors overall, and 25% in adults. This was even larger still if the drug was started early. This looked compelling, and it shaped a decade of practice.

Unfortunately, we know that this is exactly the kind of data that flatters a drug given to the people most likely to recover anyway. The patients who get oseltamivir early are, on average, the ones who present sooner, are looked after more attentively, and are less moribund. One group re-analysed similar British H1N1 data accounting for immortal time bias (taking into account the fact that you have to survive long enough to be given the drug) and the apparent benefit evaporated (hazard ratio 1.03, 95% CI 0.64–1.65). That was an important warning, but REMAP-CAP has now randomised the question properly, and the answer points the other way.

The case for adjunctive clindamycin (giving it alongside the primary antibiotics to treat infections with the bacteria Staphylococcus aureus) is compelling. Clindamycin’s mechanism of action is to inhibit protein synthesis (stop it making proteins), therefore it should switch off the toxins that drive much of the damage in severe infections. Observational research seemed to bear this out. A review of 92 cases of staphylococcal necrotising pneumonia found that regimens including an antitoxin agent were associated with lower mortality (P = 0.007), with supporting case series in PVL-positive invasive infection and severe skin and soft-tissue disease. The same logic has also long been applied to invasive Group A Strep. It is an entirely reasonable thing to have believed.

The SNAP clinical trial (for which I sit on the Global Trial Steering Committee) asked whether it actually helps in Staphylococcus aureus bloodstream infection. In this arm of the study, patients were randomised to receive 5 days Clindamycin in addition to their first line antibiotic, or not receive any adjunctive treatment. Across nearly 3,800 patients, 90-day mortality was 16.9% with clindamycin versus 15.2% without (adjusted OR 1.15), and the trial declared the intervention futile with very high probability.

But it gets worse. In the ICU subgroup, who are the sickest patients for whom the toxin argument is most seductive, mortality was 30.4% with clindamycin against 24.7% without (adjusted OR 1.47). This leaves a posterior probability >90% of increased mortality. More diarrhoea, no benefit, and a worrying signal of increased mortality precisely where we were most tempted to reach for it.

In both cases the instinct is the same, and is completely natural:

The patient is desperately sick, so do everything.

Give an antiviral. Add a second antibiotic. The mechanism makes sense, the observational data point the right way, and doing less feels uncomfortable for physicians.

But the observational data point the right way because of who receives these drugs, not because of what the drugs do. The only way to separate the two is to randomise. When we finally do, the answer is sometimes that the thing we were sure would help does nothing, or in this case, possibly worse.

We must always remember that “do everything”, without sufficient evidence, carries a high risk of doing something that actively harms the patient

This now leaves us in an awkward position. Until the full papers are published and digested, many guidelines still say to use these drugs, and we now have good reason to believe that in some patients they may do real harm1. The clinician Q&A for REMAP-CAP states that these findings don’t apply to other populations, such as children <12 years; but what are my team supposed to advise when 10 or 11 year olds are admitted to PICU with flu this winter?

This is a firm reminder that “we must do more for the sickest” is a human reflex, not an evidence base. The discipline of waiting for a trial, rather than acting on what feels right, is the thing that actually protects patients.

More to follow when both papers are out in full.