Differences between protocol and review

We revised our planned methods of assessing the risk of bias of the included trials and used the Cochrane risk of bias assessment tool 2 (RoB2 tool)³⁰ instead of the first version of the tool as stated in the protocol.²⁸ Research in the field has progressed, and RoB2 reflects current understanding of how the causes of bias may influence trial results and is the most appropriate way to assess the risk of bias. Furthermore, we revised our planned methods for confirming or rejecting a fixed intervention effect of 25% to instead use trial sequential analyses to define the lowest intervention effects threshold we could confirm or reject.

All cause mortality

Sixteen trials (19 comparisons) with a total of 2415 participants reported on all cause mortality. The included trials assessed the effects of five different fever therapy interventions: ibuprofen versus placebo (three trials),^{47 48 49} paracetamol versus placebo (six trials),^{50 51 52 53 54 55} physical cooling versus no intervention (three trials),^{56 57 58} antipyretics plus physical cooling versus no intervention (two trials),^{18 59} and physical cooling plus antipyretics versus antipyretics (two trials).^{60 61} All 2415 participants were admitted to hospital; 2050 were critically ill, and 251 were non-critically ill (supplementary table S1). One trial included both critically ill and non-critically ill patients (120/2415) (supplementary table S1). Of the 2415 participants, 1658 had infectious fever and 477 had non-infectious fever (supplementary table S1). For 286/2415 participants, the origin of fever was unknown (supplementary table S1). A total of 294 (23.0%) of 1281 fever therapy participants died compared with 258 (22.6%) of 1140 control participants.

Meta-analysis of all cause mortality did not show evidence of a difference (risk ratio 1.04, 95% confidence interval 0.90 to 1.19; I²=0%; P=0.62; 16 trials; high certainty evidence) (fig 2; supplementary table S3). Quantitative measures of heterogeneity (I²=0%) combined with visual inspection of the forest plot did not show signs of significant heterogeneity (fig 2). Trial sequential analysis showed that we could reject the hypothesis that fever therapy reduces the risk of all cause mortality by 22% (fig 3). We assessed this outcome result as being at high risk of bias and the certainty of the evidence as being high (supplementary table S3). None of the subgroup analyses showed evidence of a difference (supplementary figures S3-S10). The assessment time points varied between trials, ranging from one day after randomisation to 90 days after randomisation.^{54 56}

Open in a separate window

Fig 2

Random effects meta-analysis comparing fever therapy versus control interventions for all cause mortality (risk ratio 1.04, 95% confidence interval 0.90 to 1.19; P=0.62; I²=0%; 16 trials)

Open in a separate window

Fig 3

Trial sequential analysis of fever therapy versus control interventions for all cause mortality. Two sided trial sequential analysis graph of fever therapy versus control interventions for all cause mortality in 16 trials. Diversity adjusted required information size (DARIS) was calculated on basis of all cause mortality proportion in control group of 22.6%, relative risk reduction of 22% in experimental group, type I error (α) of 2%, and type II error (β) of 10% (90% power). Diversity was 10%. Required information size was calculated to be 3760 participants. Cumulative z curve (purple line) did not cross trial sequential monitoring boundaries for either benefit or harm. Cumulative z curve did cross inner wedge futility line (yellow outward sloping lines). Black dashed lines show conventional boundary (α=5%)

Serious adverse events

Sixteen trials (19 comparisons) with a total of 2415 participants reported on serious adverse events. The included trials assessed the effects of five different fever therapy interventions: ibuprofen versus placebo (three trials),^{47 48 49} paracetamol versus placebo (six trials),^{50 51 52 53 54 55} physical cooling versus no intervention (three trials),^{56 57 58} antipyretics plus physical cooling versus no intervention (two trials),^{18 59} and physical cooling plus antipyretics versus antipyretics (two trials).^{60 61} All 2415 participants were admitted to hospital; 2050 were critically ill, and 251 were non-critically ill (supplementary table S1). One trial included both critically ill and non-critically ill patients (120/2415). Of the 2415 participants, 1658 had infectious fever and 477 had non-infectious fever (supplementary table S1). For 286/2415 participants the origin of fever was unknown (supplementary table S1). A total of 307 (24.0%) of 1281 trial participants had a serious adverse event in the fever therapy group compared with 276 (24.2%) of 1140 in the control group.

Meta-analysis of serious adverse events did not show evidence of a difference (risk ratio 1.02, 0.89 to 1.17; I²=0%; P=0.78; 16 trials; high certainty evidence) (fig 4; supplementary table S3). Quantitative measures of heterogeneity (I²=0%) combined with visual inspection of the forest plot did not show signs of significant heterogeneity (fig 4). Trial sequential analysis showed that we could reject the hypothesis that fever therapy reduces the relative risk of serious adverse events by 23% (fig 5). We assessed this outcome result as being at high risk of bias and the certainty of the evidence as being high (supplementary table S3). None of the subgroup analyses showed evidence of a difference (supplementary figures S12-S19). The assessment time points varied between trials, ranging from one day after randomisation to 90 days after randomisation.^{54 56}

Open in a separate window

Fig 4

Random effects meta-analysis comparing fever therapy versus control interventions for serious adverse events (risk ratio 1.02, 95% confidence interval 0.89 to 1.17; P=0.78; I²=0%; 16 trials)

Open in a separate window

Fig 5

Trial sequential analysis of fever therapy versus control interventions for serious adverse events. Two sided trial sequential analysis graph of fever control interventions versus control interventions for serious adverse events in 16 trials. Diversity adjusted required information size (DARIS) was calculated on basis of all cause mortality proportion in control group of 22.6%, relative risk reduction of 23% in experimental group, type I error (α) of 2%, and type II error (β) of 10% (90% power). Diversity was 21%. Required information size was calculated to be 3916 participants. Cumulative Z curve (purple line) did not cross trial sequential monitoring boundaries for either benefit or harm. Cumulative Z curve did cross inner wedge futility line (yellow outward sloping lines). Black dashed lines show conventional boundary (α=5%).

Type of fever therapy

In this systematic review, we chose to include both antipyretic drugs and physical cooling when assessing the effects of fever therapy. The physiological effects and underlying mechanisms of action of antipyretic drugs and physical cooling are different. Antipyretic drugs mainly affect body temperature by lowering the thermal setpoint and are reliant on a functioning thermoregulatory centre to have an effect.⁶² Physical cooling does not affect the thermoregulatory centre and acts by removing heat from the body without affecting the hypothalamic setpoint; the cooling is thus forced on the body, which may cause adverse effects. Decreasing the core body temperature without affecting the hypothalamic setpoint induces cold defensive responses mediated via the sympathetic nervous system such as shivering, tachycardia, and peripheral vasoconstriction, which lead to discomfort for the patient and increased metabolic stress.⁶³ The compensatory responses to the discrepancy between the body temperature and the thermal setpoint generate and retain heat, counteracting the cooling and increasing the metabolic demand. These cold defensive responses may be reduced pharmacologically with muscle paralysing drugs, analgesics, and sedatives, which in turn may lead to additional adverse effects. Antipyretics decrease the setpoint and therefore do not induce cold defensive responses or affect the metabolic demand.⁶⁴

The different mechanisms of action and adverse effects have spawned theories that specific approaches may be tailored to patients with different origins of fever to adapt a treatment regimen to achieve superior results. Physical cooling is theorised to be more appropriate than antipyretics for patients with acute brain injury, as the thermoregulatory centre could be compromised,⁶⁵ and antipyretics could be more appropriate in patients with infectious fever.⁶⁶ Clinically, however, antipyretic refractory infectious fever is commonly observed, raising questions about the validity of these theories. Most trials analysed in the meta-analyses and trial sequential analyses investigated antipyretic drugs (9/16; 56%) rather than physical cooling (5/16; 31%), and only two trials investigated a combination of antipyretics and physical cooling (2/16; 13%). Hence, even though we did not identify signs of heterogeneity, our results primarily show the effects of antipyretic drugs, which needs to be considered when interpreting our results.

Strengths and limitations in relation to other studies

Previous studies have primarily focused on specific patient groups, such as critically ill patients and those with sepsis, or specific fever control interventions.^{20 21 22 23 24} Our review differs from previous research in that we have included a wider variety of both participants and fever therapies, allowing us, unlike previous studies, to obtain sufficient power by reaching required information sizes. The implications of including a variety of different interventions and a diverse patient group may entail results that are difficult to interpret if significant statistical heterogeneity is present. We therefore stated in the protocol that meta-analyses would be done only if the trials showed low levels of heterogeneity. With our analyses indicating low levels of heterogeneity and our subgroup analyses showing no subgroup differences, we deemed meta-analyses to be valid and did them including all identified trials. The relative effect, regardless of intervention type and patient group, being similar confirms and verifies our decision to pool these trials. We do, however, acknowledge that tests of heterogeneity most likely are underpowered.

The scope of this systematic review was wider in the population studied compared with previous research because our ambition was to assess the effects of fever therapy with regard to the outcomes most important to patients—that is, mortality and serious adverse events. Trials designed to investigate and report on these outcomes will inevitably focus on sicker patient populations; the participants included in the important analyses were all admitted to hospital, and most were critically ill. Our results therefore say little about fever therapy for milder disease states in the wider non-admitted population.

Biological rationale

Whether aiming for a benefit in terms of mortality and morbidity with fever therapy is biologically and physiologically plausible is questionable. From a phylogenetic perspective, with the conserved fever response, one might expect fever to be beneficial and that fever therapy should be avoided. However, that the physiological stress caused by fever shifts the balance in a vulnerable and critically ill patient towards worse outcomes is equally plausible.⁶⁷ Nevertheless, different types of fever therapy have been part of usual care across medical specialties for decades. Results of non-randomised studies are conflicting, and the results from our analyses do not indicate any effects of the interventions studied. One important finding from this study is that the total sample size of participants included in randomised trials was relatively small and that less than half of the trials included reported on hard outcomes. We investigated the smallest intervention effects we could reject using trial sequential analyses: 22% for mortality and 23% for serious adverse events. We recognise that smaller intervention effects may be beneficial for the affected patients and thus clinically relevant. Our results indicate that further trials should therefore be powered for detecting smaller and more realistic intervention effects.

Strengths and limitations of study

Our review has several strengths. Our method was predefined in detail and published before we did our literature search. We searched all relevant databases, we used an eight step assessment suggested by Jakobsen and colleagues to assess the clinical significance of our results, and we used trial sequential analysis to reduce the risks of type I and type II errors. Furthermore, we did meta-analyses with both fixed effects and random effects meta-analysis, we investigated subgroup differences, and we assessed the certainty of the evidence through GRADE. The main limitation of our review was the low methodological quality of the included trials, with half of the included trials assessed as being of some concerns and half assessed as being at high risk of bias. The inclusion of active comparator trials (that is, physical cooling plus antipyretics versus antipyretics, antipyretics plus physical cooling versus physical cooling, or aggressive treatment versus standard treatment) along with experimental versus control trials is a potential complicating factor with regard to the interpretation of the results. However, in the absence of heterogeneity between trials, as in our study, this should not be considered limiting to our results. Aiming to be inclusive, we accepted a wide variety of patients and interventions. Despite no heterogeneity being seen, specific cases may exist in which fever therapy is beneficial or harmful.

Meaning of study

Antipyretics are routinely administered in care both in and out of hospital, and physical cooling is applied in severe fever not responding to antipyretic drugs. Our results suggest little to no difference in effect between fever therapy and no fever therapy; realistic, while still clinically relevant, intervention effects of fever therapy have so far not been properly studied, however. Fever therapy may be investigated in further adequately powered high quality randomised trials also including a health economics perspective, to define implications for patients and society.