Of 8,455 citations identified in the initial search, we assessed 212 full texts for eligibility (Figure 1). Of these, 44 were deemed eligible and included in the synthesis (Additional file 3). Details of excluded systematic reviews are provided in Additional file 4.
Description of included reviews
At least one systematic review was identified for ten out of 14 targeted low-value practices (Table 1). Systematic reviews conducted within the last five years were available for nine clinical practices and meta-analyses were available for eight practices. The number of systematic reviews varied from one for plasma transfusion and neuromuscular blocking agents to 19 for hypothermia. Reviews on imaging mostly defined primary outcomes as intracranial injury, neurological deterioration or neurosurgical intervention, and their population as patients with mild or mild complicated TBI. Systematic reviews on therapeutic interventions mainly focussed on the GOS or GOS-Extended, mortality, or adverse events in patients with moderate to severe TBI. Most systematic reviews restricted their population of interest to adults but some included pediatric patients. Nine systematic reviews did not specify targeted study designs in their PICOS. In 20 other systematic reviews, only RCTs were included and in 14 both randomized controlled trials and observational studies were considered.
Methodological quality of systematic reviews
Of the 44 included systematic reviews, two[41,42] were rated high quality and eight[43-50] moderate quality (Additional file 5). All but two reviews[51,52][51, 52] used a comprehensive research strategy (95%), 17 (38%) established methods prior to the review and reported significant deviations from the study protocol, 34 (76%) used a satisfactory technique for reporting risk of bias, 20 (45%) accounted for risk of bias in individual studies when interpreting/discussing results, 33 out of the 34 reviews (97%) performing meta-analyses used appropriate analytic methods, and 20 (59%) investigated the presence of publication bias and discussed its potential impact on the results of meta-analyses.
Synthesis of results
Primary outcomes
Diagnostic interventions
We identified two reviews for CT in adults with mild TBI (both without meta-analyses),[44,53] but only one presented data allowing us to calculate point estimates for our primary outcome;[44] less than 5% of patients who were classed as low-risk (any decision rule) had intracranial injury (Table 2). Sample sizes were large for studies using the Canadian CT Head Rule (CCHR), the CT in Head Injury Patients (CHIP) rule and National Emergency X-Radiography Utilization Study (NEXUS) rule, but none of the included studies were at low risk of bias. In adults with acute mild complicated TBI (abnormal initial CT) with no neurological deterioration, routine repeat CT detected progression of intracranial hemorrhage in around 20% of patients. Routine repeat head CT led to the detection of delayed intracranial in only 0.6% of adults with mild TBI on anticoagulant or antiplatelet therapy.[54]
Therapeutic interventions
We identified two systematic reviews with meta-analyses on platelet transfusion in adults with TBI on antiplatelet therapy, which suggested increased risk of mortality in patients receiving the intervention, but estimates were imprecise and all CIs included the null value.[46,55] The systematic review on antibiotic prophylaxis for basal skull fracture suggested that the intervention is associated with reduced odds of meningitis, but again the estimates were imprecise and covered the null value.[47] A systematic review by the same group reported no benefit of antibiotic prophylaxis for external ventricular drain placement in severe TBI in terms of risk of infection.[41] We identified four systematic reviews with meta-analyses on seizure prophylaxis extended for more than one week after injury.[56,57,48,58] Odds ratio (OR) and risks ratio (RR) varied from 0.40 to 1.28 across systematic reviews. The most widely studied drug, levetiracetam, was associated with a potential reduction in late seizures, but with confidence intervals (CI) covering the null value. Fifteen out of the nineteen systematic reviews identified for therapeutic hypothermia performed meta-analyses with the GOS as the primary outcome.[59-64,43,65,42,66,49,67,52,68,69] OR/RRs varied between 0.61 and 1.16 with 11 suggesting significant benefit (credibility of evidence 3 class II,[43,42,69] three class III[65,67,68] and 5 class IV[61-64,42]) and one[69] (the most recent) suggesting significant harm (class IV). In five meta-analyses on high-quality studies, OR/RR either covered the null value (n=3),[64,65,49] suggested significant harm (n=1)[69] or suggested significant benefit (n=1).[42] Finally, we identified six systematic reviews on decompressive craniectomy,[70-75] of which four presented quantitative synthesis on GOS.[70,71,74,75] All were based exclusively on RCTs but were of low or critically low quality and had highly heterogeneous point estimates (I2≥72%). Effect estimates were consistently close to one with CI covering the null value, suggesting no significant difference in outcome between intervention and control groups.
Secondary outcomes
Close to 0% of adults with mild TBI who were at low risk on a clinical decision rule for head CT required neurosurgical intervention (Additional file 6).[44] Routine repeat head CT in mild complicated TBI without neurological deterioration led to a neurosurgical intervention in between 0.6 and 2.4% of patients,[76-78] a change in clinical management in between 0.6[77] and 3.9%[78] and a change in intracranial pressure (ICP) monitoring in 1.2%.[78] Less than 0.2% of adults with mild TBI on anticoagulant or antiplatelet therapy with a routine repeat CT required neurosurgical treatment or died in hospital.[54]
Authors of the single systematic review on antibiotic prophylaxis in adults with basal skull fractures reported no reduction in all-cause or meningitis-related mortality.[47] In, eight of the 15 systematic reviews on hypothermia, a statistically significant reduction in the risk or odds of mortality was observed whereas one review observed an increase.[61,64,43,65,67,68,50,69] Five out of seven reviews that looked at pneumonia suggested higher risk/odds of this adverse event in adults receiving therapeutic hypothermia.[59,64,49,67,69] All of the four systematic reviews on decompressive craniectomy looked at mortality;[71,73-75] three observed a statistically lower risk in the intervention group at 6 months.[73-75] Finally, two systematic reviews on decompressive craniectomy observed significantly lower mean intracranial pressure and shorter length of stay in the intervention group but significantly higher risk/odds of complications.[73,75]