Hope: The Mother of Bias in Research

I realized the other day that underlying every slanted report or overly-optimistic interpretation of a trial's results, every contorted post hoc analysis, every Big Pharma obfuscation, is hope.  And while hope is generally a good, positive emotion, it engenders great bias in the interpretation of medical research.  Consider this NYT article from last month:  "Dashing Hopes, Study Shows Cholesterol Drug Had No Effect on Heart Health."  The title itself reinforces my point, as do several quotes in the article.

“All of us would have put money on it,” said Dr. Peter Libby, a Harvard cardiologist. The drug, he said, “was the great hope.”

 Again, hope is wonderful, but it blinds people to the truth in everyday life and I'm afraid researchers are no more immune to its effects than the laity.  In my estimation, three main categories of hope creep into the evaluation of research and foments bias:

1. Hope for a cure, prevention, or treatment for a disease (on the part of patients, investigators, or both)
2. Hope for career advancement, funding, notoriety, being right (on the part of investigators) and related sunk cost bias
3. Hope for financial gain (usually on the part of Big Pharma and related industrial interests)

Consider prone positioning for ARDS.  For over 20 years, investigators have hoped that prone positioning improves not only oxygenation but also outcomes (mostly mortality).  So is it any wonder that after the most recent trial, in spite of the 4 or 5 previous failed trials, the community enthusiastically declared "success!"  "Prone Positioning works!"  Of course it is no wonder - this has been the hope for decades.

But consider what the most recent trial represents through the lens of replicability:  a failure to replicate previous results showing that prone positioning does not improve mortality.  The recent trial is the outlier.  It is the "false positive" rather than the previous trials being the "false negatives."

This way of interpreting the trials of prone positioning in the aggregate should be an obvious one, and it astonishes me that it took me so long to see the results this way - as a single failure to replicate previously replicable negative results.  But it hearkens to the underlying bias - we view results through the magnifying glass of hope, and it distorts our appraisal of the evidence.

Indeed, I have been accused of being a nihilist because of my views on this blog, which some see as derogating the work of others or an attempt to dash their hopes.  But these critics engage, or wish me to engage in a form of outcome bias - the value of the research lies in the integrity of its design, conduct, analysis, and reporting, not in its results.  One can do superlative research and get negative results, or shoddy research and get positive results.  My goal here is and always has been to judge the research on its merits, regardless of the results or the hopes that impel it.

(Aside:  Cholesterol researchers have a faith or hope in the cholesterol hypothesis - that cholesterol is a causal factor in pathways to cardiovascular outcomes.  Statin data corroborate this, and preliminary PCSK9 inhibitor data do, too.  But how quickly we engage in hopeful confirmation bias!  If cholesterol is a causal factor, it should not matter how you manipulate it - lower the cholesterol, lower cardiovascular events.  The fact that it does appear to matter how you lower it suggests that either there are multiplicity of agent effects (untoward and unknown effects of some agents negate some their beneficial effects in the cholesterol causal pathway) or that cholesterol levels are epiphenomena - markers of the effects of statins and PCSK9 inhibitors on the real, but as yet undelineated causal pathways.  Maybe the fact that we can easily measure cholesterol and that it is associated with outcomes in untreated individuals is a convenient accident of history that led us to trial statins which work in ways that we do not yet understand.)

Much Ado About Nothing? The Relevance of New Sepsis Definitions for Clinical Care Versus Research

What's in a name?  That which we call a rose, by any other name would smell as sweet. - Shakespeare, Romeo and Juliet Act II Scene II

The Society of Critical Care Medicine is meeting this week, JAMA devoted an entire issue to sepsis and critical illness, and my twitter feed is ablaze with news of release of a new consensus definition of sepsis.  Much laudable work has been done to get to this point, even as the work is already generating controversy (Is this a "first world" definition that will be forced upon second and third world countries where it may have less external validity?  Why were no women on the panel?).  Making the definition of sepsis more reliable, from a sensitivity and specificity standpoint (more accurate) is a step forward for the sepsis research enterprise, for it will allow improved targeting of inclusion criteria for trials of therapies for sepsis, and better external validity when those therapies are later applied in a population that resembles those enrolled.  But what impact will/should the new definition have on clinical care?  Are the-times-they-are-a-changing?

Diagnosis, a fundamental goal of clinical medicine is important for several reasons, chief among them:

1. To identify the underlying cause of symptoms and signs so that treatments specific to that illness can be administered
2. To provide information on prognosis, natural history, course, etc for patients with or without treatment
3. To reassure the physician and patients that there is an understanding of what is going on; information itself has value even if it is not actionable

Thus redefining sepsis (or even defining it in the first place) is valuable if it allows us to institute treatments that would not otherwise be instituted, or provides prognostic or other information that is valuable to patients.  Does it do either of those two things?

Is There a Baby in That Bathwater? Status Quo Bias in Evidence Appraisal in Critical Care

"But we are not here concerned with hopes and fears, only the truth so far as our reason allows us to discover it."  -  Charles Darwin, The Descent of Man

Status quo bias is a cognitive decision making bias that leads to decision makers' preference for the choice represented by the current status quo, even when the status quo is arbitrary or irrelevant.  Decision makers tend to perceive a change from the status quo as a loss and therefore their decisions are biased toward the status quo.  This can lead to preference reversals when the status quo reference frame is changed.  The status quo can be debiased using a reversal test, i.e., manipulating the status quo either experimentally or via thought experiment to consider a change in the opposite direction.  If reluctance to change from the status quo exists in both directions, status quo bias is likely to exist.

My collaborators Peter Terry, Hal Arkes and I reported in a study published in 2006 that physicians were far more likely to abandon a therapy that was status quo or standard therapy based on new evidence of harm than they were to adopt an identical therapy based on the same evidence of benefit from a fictitious RCT (randomized controlled trial) presented in the vignette.  These results suggested that there was an asymmetric status quo bias - physicians showed a strong preference for the status quo in the adoption of new therapies, but a strong preference for abandoning the status quo when a standard of care was shown to be harmful.  Two characteristics of the vignettes used in this intersubject study deserve attention.  First, the vignettes described a standard or status quo therapy that had no support from RCTs prior to the fictitious one described in the vignette.  Second, this study was driven in part by what I perceived at the time was a curious lack of adoption of drotrecogin-alfa (Xigris), with its then purported mortality benefit and associated bleeding risk.  Thus, our vignettes had very significant trade-offs in terms of side effects in both the adopt and abandon reference frames.  Our results seemed to explain s/low uptake of Xigris, and were also consistent with the relatively rapid abandonment of hormone replacement therapy (HRT) after publication of the WHI, the first RCT of HRT.

Enrolling Bad Patients After Good: Sunk Cost Bias and the Meta-Analytic Futility Stopping Rule

Four (relatively) large critical care randomized controlled trials were published early in the NEJM in the last week.  I was excited to blog on them, but then I realized they're all four old news, so there's nothing to blog about.  But alas, the fact that there is no news is the news.

In the last week, we "learned" that more transfusion is not helpful in septic shock, that EGDT (the ARISE trial) is not beneficial in sepsis, that simvastatin (HARP-2 trial) is not beneficial in ARDS, and that parental administration of nutrition is not superior to enteral administration in critical illness.  Any of that sound familiar?

I read the first two articles, then discovered the last two and I said to myself "I'm not reading these."  At first I felt bad about this decision, but then that I realized it is a rational one.  Here's why.

Sell Side Bias and Scientific Stockholm Syndrome: A Report from the Annual Meeting of the American Thoracic Society

What secrets lie inside?

Analysts working on Wall Street are sometimes categorized as working on either the "buy side" or the "sell side" depending on whether their firm is placing orders for stocks (buy side, such as institutional investors for mutual funds) or filling orders for stocks (sell side, which makes commissions on stock trades).  Sell side bias refers to any tendency for the sell side to "push" stocks via overly optimistic ratings and analyses.

Well, I'm at the American Thoracic Society (ATS) meeting in San Diego right now, and it certainly does feel like people - everyone - is trying to sell me something.  From the giant industry sponsored banners, to the emblazoned tote bags, to the bags of propaganda left at my hotel room door every morning, to the exhibitor hall filled with every manner of new and fancy gadgets (but closed to cameras), to the investigators themselves, everybody is trying to convince me to buy (or prescribe) something.  Especially ideas.  Investigators have a promotional interest in their ideas.  And they want you and me to buy into their ideas.  I have become convinced that investigators without industry ties (that dying breed) are just about as susceptible to sell side bias as those with industry ties.  Indeed, I have also noted that the potential consumer of many of the ideas himself seems biased - he wants things to work, too, and he has a ready explanation for why some ideas didn't pan out in the data (see below).  It's like an epidemic of scientific Stockholm Syndrome.

The first session I attended was a synopsis of the SAILS trial by the ARDSnet investigators, testing whether use of a statin, rosuvastatin, in patients with sepsis-incited lung injury would influence 60 day mortality.  The basis of this trial was formed by observational associations that patients on statins had better outcomes in this, that, and the other thing, including sepsis.  If you are not already aware of the results, guess whether rosuvastatin was beneficial in this study.

Over Easy? Trials of Prone Positioning in ARDS

Published May 20 in the  NEJM to coincide with the ATS meeting is the (latest) Guerin et al study of Prone Positioning in ARDS.  The editorialist was impressed.  He thinks that we should start proning patients similar to those in the study.  Indeed, the study results are impressive:  a 16.8% absolute reduction in mortality between the study groups with a corresponding P-value of less than 0.001.  But before we switch our tastes from sunny side up to over easy (or in some cases, over hard - referred to as the "turn of death" in ICU vernacular) we should consider some general principles as well as about a decade of other studies of prone positioning in ARDS.

First, a general principle:  regression to the mean.  Few, if any, therapies in critical care (or in medicine in general) confer a mortality benefit this large.  I refer the reader (again) to our study of delta inflation which tabulated over 30 critical care trials in the top 5 medical journals over 10 years and showed that few critical care trials show mortality deltas (absolute mortality differences) greater than 10%.   Almost all those that do are later refuted.  Indeed it was our conclusion that searching for deltas greater than or equal to 10% is akin to a fool's errand, so unlikely is the probability of finding such a difference.  Jimmy T. Sylvester, my attending at JHH in late 2001 had already recognized this.  When the now infamous sentinel trail of intensive insulin therapy (IIT) was published, we discussed it at our ICU pre-rounds lecture and he said something like "Either these data are faked, or this is revolutionary."  We now know that there was no revolution (although many ICUs continue to practice as if there had been one).  He could have just as easily said that this is an anomaly that will regress to the mean, that there is inherent bias in this study, or that "trials stopped early for benefit...."

HFOV Fails as a Routine Therapy for moderate-to-severe ARDS. Musings on the Use and Study of “Rescue Therapies”.

Ferguson et al report the results of the OSCILLATE randomized controlled trial of HFOV for moderate to severe ARDS in this week’s NEJM.  (A similar RCT of HFOV, the OSCAR trial, is reported in the same issue but I limit most of my commentary to OSCILLATE because I think it’s a better and more interesting trial and more data are presented in its report.)  A major question is answered by this trial, but an important question remains open:  is HFOV an acceptable and rational option as “rescue therapy” in certain patients with “refractory” ARDS?  I remain undecided about this question, and its implications are the subject of this post.

Before I segue to the issue of the study and efficacy of rescue therapies, let’s consider some nuances of this trial:

·         Patients in both groups received high doses of sedatives (average midazolam dose for the first week: 8.3 mg/hour in the HFOV group versus 5.9 mg/hour in the control group – a 41% increase in HFOV).  Was this “too much” sedation?  What if propofol had been used instead?

·         Patients in the HFOV group received significantly more paralytics.  If you believe the Papazian data (I don’t) paralytics should confer a mortality benefit in early ARDS and this should contribute to LOWER mortality in the HFOV group.  What if paralytics had been used less frequently?

·         Does HFOV confer a nocebo effect by virtue of its “unnatural” pattern of ventilation, its “requirement” for more sedation and paralysis, or the noise associated with its provision, or its influence on the perceptions of caregivers and patient’s families (recognizing that deaths after withdrawal of life support were similar in HFOV versus conventional ventilation (55 versus 49%, P=0.12)?

·         The respiratory frequency in the HFOV group (5.5 Hz) was at the low end of the usual range (3-15 Hz).  If a higher frequency (and a lower tidal volume) had been delivered, would the result have changed?  (Probably not.)

·         What about the high plateau pressure in the control group (32 cm H2O) despite the low tidal volume of 6.1 ml/kg PBW?  Why was not tidal volume reduced such that plateau pressure was lower than the commonly recommended target of 30 cm H2O?  Did this make a difference?  (Probably not.)

·         Why was mortality higher in the minority (12%) of control patients who were changed to HFOV (71% mortality)?  Is this related to confounding by indication or reflective of the general harmful effects of HFOV?

·         Why was there a difference between the OSCILLATE study and the OSCAR study, reported in the same issue, in terms of mortality?  Because OSCILLATE patients were sicker?  Because OSCAR control patients received higher tidal volumes, thereby curtailing the advantage of conventional ventilation?  I find this last explanation somewhat compelling.