Therapeutic hypothermia (TH) has been used to treat post-hypoxic brain injury after cardiac arrest (CA) since the late 1950s. In 2002, two landmark prospective, randomized controlled trials (RCTs) confirmed the efficacy of TH for this indication [1, 2]. An 11-center trial in Europe reported 16% absolute improvement in outcome in patients with witnessed ventricular tachycardia/ventricular fibrillation (VT/VF) arrest with use of TH [1], and a four-center Australian study found 23% improvement [2]. More than 40 non-randomized studies have reported improved outcomes with TH [3]. A 5,317-patient registry in The Netherlands noted a 6.6% drop in mortality with TH implementation [4]. A Scandinavian registry with 986 patients reported 61% survival in witnessed VT/VF arrest, 92% with good neurological outcome [5]. A meta-analysis concluded that six patients had to be treated to achieve one additional case of good outcome [6]. On these bases, professional societies began recommending the use of TH in selected patients with CA [7, 8]. A Cochrane Review supported these guidelines and conclusions [9]. Further supporting evidence came from the field of neonatology, in which seven RCTs showed improved outcomes with TH in newborns with hypoxic injury [3].

However, this evidence has been challenged [10]. The largest RCT had no strict temperature management in controls, who had an average temperature of 37.8°C [1]. Other criticisms included a lack of prior power calculation and a low rate of enrollment [10]. The second RCT successfully maintained normothermia in controls, but random assignment was by day of the month rather than per patient [2]. The critics stimulated the conduct of another, larger RCT, enrolling patients with witnessed CA regardless of initial rhythm, with predefined subgroup analyses [11]. This study enrolled 939 patients, who were cooled to either 33°C or 36°C. The results were unequivocally negative.

These findings completely contradict those of all previous studies. Should we accept the results because the study was well designed and larger than previous trials?

First, some important issues need to be clarified. TH was already the standard of care in participating hospitals, and the default option for patients not enrolled in the trial was TH. Indeed, many centers had published CA outcome data that were far better than those of this study [5, 12, 13]. Hence, admitting physicians might subconsciously have selected patients with the potential to benefit from receiving ‘routine’ TH rather than have screened them for trial eligibility. The Methods section [11] suggests that all patients were evaluated, but this seems questionable: 1,431 patients were screened, and 939 were enrolled; that is an unusually high enrollment rate of 66%. The study took place in 36 intensive care units in just over 2 years, and this translates to 18 patients screened and 12 enrolled per center per year, or one patient per center per month. This number seems extremely low.

Other potential problems include a rapid rate of active re-warming, from 33°C to 36°C in 6 hours, faster than in all previous trials; this can negate the benefits of TH [14, 15]. The temperature graph, Figure 1 in the article [11], shows wide error bars, potentially indicating large temperature swings that can be harmful [14, 15]. Also, it appears that many ‘favorable’ factors such as bystander-witnessed arrest and ‘shockable’ rhythm were more prevalent in the 36°C group but that ‘unfavorable’ factors such as circulatory shock and absence of pupillary and corneal reflexes were more common in the 33°C group [11]. The differences are small but may be cumulative. There was a greater prevalence of spontaneous hypothermia (before start of active cooling) in the 33°C group, potentially indicating greater severity of brain injury with diminished shivering response [14, 15]. There were more seizures in the 33°C group, in spite of the well-recognized anti-seizure effects of hypothermia [3]. More patients in the 33°C group met criteria for early withdrawal of care, again suggesting greater severity of injury [11].

The consequences of accepting these conclusions are momentous. We urge our colleagues not to abandon TH in favor of strict fever management (or, worse, no temperature management) on the basis of one study, until all relevant issues have been satisfactorily addressed.