Kjetil Gundro Brurberg

Case definitions for chronic fatigue syndrome/myalgic encephalomyelitis (CFS/ME): a systematic review

Objective To identify case definitions for chronic fatigue syndrome/myalgic encephalomyelitis (CFS/ME), and explore how the validity of case definitions can be evaluated in the absence of a reference standard. Design Systematic review. Setting International. Participants A literature search, updated as of November 2013, led to the identification of 20 case definitions and inclusion of 38 validation studies. Primary and secondary outcome measure Validation studies were assessed for risk of bias and categorised according to three validation models: (1) independent application of several case definitions on the same population, (2) sequential application of different case definitions on patients diagnosed with CFS/ME with one set of diagnostic criteria or (3) comparison of prevalence estimates from different case definitions applied on different populations. Results A total of 38 studies contributed data of sufficient quality and consistency for evaluation of validity, with CDC-1994/Fukuda as the most frequently applied case definition. No study rigorously assessed the reproducibility or feasibility of case definitions. Validation studies were small with methodological weaknesses and inconsistent results. No empirical data indicated that any case definition specifically identified patients with a neuroimmunological condition. Conclusions Classification of patients according to severity and symptom patterns, aiming to predict prognosis or effectiveness of therapy, seems useful. Development of further case definitions of CFS/ME should be given a low priority. Consistency in research can be achieved by applying diagnostic criteria that have been subjected to systematic evaluation.

ST waveform analysis vs. cardiotocography alone for intrapartum fetal monitoring: A systematic review and meta-analysis of randomized trials

ST waveform analysis was introduced to reduce metabolic acidosis at birth and avoid unnecessary operative deliveries relative to conventional cardiotocography. Our objective was to quantify the efficacy of ST waveform analysis vs. cardiotocography and assess the quality of the evidence using the Grading of Recommendations Assessment, Development and Evaluation tool.

Use of forms for follow-up of diabetes in general practice.

BACKGROUND The majority of patients with type 2 diabetes are followed up in general practice. We have investigated whether the use of forms by GPs for recording clinical data contributes to lower mortality and morbidity for this patient group. MATERIAL AND METHOD This systematic review is based on literature searches in MEDLINE, EMBASE, ISI Web of Science, Cochrane CENTRAL and PubMed. We included studies that 1) dealt with adults over 18 years of age with diabetes who were followed up in the primary health service and 2) compared mortality and morbidity with and without the use of forms. We summarised the results qualitatively and using meta-analyses. RESULTS Seven studies were included. One study (1262 participants) investigated the effect of the form on hard endpoints, without finding clear effects on mortality (HR 0.91; 95% CI 0.72-1.14), retinopathy (OR 0.90; 95% CI 0.53-1.52), peripheral nerve injury (OR 0.86; 95% CI 0.57-1.29), myocardial infarction (OR 0.65; 95% CI 0.31-1.35) or stroke (OR 0.89; 95% CI 0.39-2.01). Use of forms appears to have little or no effect on body weight (3 studies), and a small, positive effect on blood pressure (5 studies) and total cholesterol (2 studies). INTERPRETATION Published data at present provide no clear answers, but shows that use of forms in the follow-up of patients with diabetes in general practice may tend to contribute to lower mortality and morbidity.

STAN technology, surrogate outcomes and possible sources of bias

Sir, We appreciate the comments from Olofsson (1) and Kessler et al. (2). We investigated several outcomes in our review, and two (risk of metabolic acidosis and operative vaginal delivery) showed statistically significant differences in favor of STAN. (3). Two other high quality meta-analyses were published almost simultaneously (4,5), and neither reported significant differences in rates of metabolic acidosis. Saccone et al. (4) and Neilson (5) used risk ratio (RR) and random effect models, whereas we used peto odds ratio (OR) and fixed effect model for outcomes with an incidence <1%. Both approaches have pros and cons and we hesitate to define either one as superior. Olofsson argues that we did the most correct metaanalysis (3). Positive feedback is always welcome, but it is tempting to ask whether this judgment is related to the fact that our metaanalysis shows positive results for metabolic acidosis, whereas the others do not. We once again emphasize the need to view all results in context, particularly when the only interesting effect manifests itself in a surrogate outcome with uncertain clinical validity. Surrogate outcomes are used to predict the risk of future serious events, thus shortening the size, duration and cost of trials. Unfortunately, this is associated with pitfalls and bias (6,7). The uncertain validity of metabolic acidosis is demonstrated in an individual patient data review (4) investigating a composite endpoint (at least one of the following: intrapartum fetal death, neonatal death, Apgar score ≤3 at five minutes, neonatal seizures, metabolic acidosis, intubation for ventilation at delivery or neonatal encephalopathy) without finding a difference between STAN and CTG. Hence, we disagree with Kessler et al. and Olofsson, who seem to take the validity of metabolic acidosis for granted. Kessler et al. assume that 10.3% of all babies born with metabolic acidosis have severe adverse outcomes due to an intrapartum hypoxic event. Their calculation presupposes that the risk reductions for metabolic acidosis and for serious adverse events are linearly related. We find this inference speculative, and wonder why one should trust estimates based on assumption rather than direct data. Direct data suggest that that STAN might be associated with reduced survival (3–5). Kessler et al. estimated that STAN will prevent 493 operative vaginal deliveries in Norway each year. This estimate is based on the questionable assumption that all delivery units use STAN on all laboring women, and they ask whether we regard this reduction as unimportant. We welcome efforts to reduce operative deliveries without compromising neonatal outcomes, but this should involve other approaches rather than STAN. Olofsson further argues that our GRADE assessments are influenced by culture, norms and other preferences. This is of course true. The use of GRADE does not guarantee consensus, but we note that the Cochrane meta-analysis (5) arrived at very similar conclusions. Olofsson states that relying more on negative than positive evidence is a part of being human, suggesting that our conclusions are prone to bias. We believe that confirmation bias, conflicts of interest and uncritical embracement of new technology (8) are the most potent sources of bias in this field. We do not see any reasons why we could be more exposed to this type of bias than others. As an extension of the latter argument, it is tempting to refer to the criticism of the recently published US study (9). This study was funded with 3 million USD and supported by Neoventa AB (10). The STAN algorithm was different from that used in Europe, but the algorithm was the same as used by Neoventa for their FDA approval (10). In 2014, Olofsson, Kessler, Yli and others published a review (11) and concluded: “The results of the ongoing multicenter RCT in the United States are some months away. Certainly the contribution of the USA data will help to determine whether the addition of ST analysis to conventional CTG results in improved perinatal outcomes.” After the US study showed negative results, Kessler, Yli and others published a statement with severe objections to this study on the Neoventa homepage (12). It is tempting to speculate whether this criticism would have been raised if the US study had published positive results.

Statistical significance is not necessarily equal to clinical significance.

Sir, We thank Vayssiere et al. for their comment (1). Vayssiere et al. state that there were contradictory findings regarding neonatal metabolic acidosis between the three recently published meta-analyses on STAN vs. CTG alone (2–4). We disagree, as all three meta-analyses reported a decrease but only one reached statistical significance. Further, Vayssiere et al. state that only we used revised data from previous trials appropriately, and they have performed new analyses using different methods based on the numbers used in our study (2). All their new analyses reached statistical significance, and Vayssiere et al. conclude that that the STAN method reduces metabolic acidosis by one-third and conclude that it is beneficial. We reported a statistically significant difference (relative risk reduction 36%, absolute risk reduction 0.25%). The significance was lost when we used peto OR in combination with a randomeffect model rather than a fixed-effect model. We argued that the result should be interpreted with caution because metabolic acidosis is a surrogate endpoint with a questionable relation to harder outcomes and because we did not observe similar effects in other clinical outcomes. In addition, the statistically significant result disappeared when using another method, which underpins the need for caution. We have previously argued carefully for our view (2,5) and will not repeat the arguments here. Vayssiere et al. (1) state that metabolic acidosis is one of the best indicators available at birth for the immediate assessment – without providing any arguments as to why it makes metabolic acidosis a predictive indicator for adverse fetal outcomes in this setting. If they could provide studies showing good correlations between metabolic acidosis and adverse outcomes, it would be useful. No long-term outcomes, as cerebral palsy, are published from the randomized controlled trials using the STAN technology. After publishing our meta-analysis, we have experienced that supporters of STAN conclude that we used the most appropriate methods and the right data – but criticized us for our conclusions. They are unwilling to discuss the problem about use of surrogate outcomes and risk of bias (6,7), or the low absolute risk reduction of 0.25%. Based on data from six trials of good quality, with more than 26 000 women included, there is currently no evidence to conclude that STAN improves neonatal outcomes compared with CTG alone. We are now discussing whether small changes in numbers extracted from the primary studies or slightly different methods can change results and conclusions in the three recent meta-analyses. It is unlikely that future trials comparing STAN with CTG alone will be funded, and it is unlikely that evidence from observational studies can change the conclusions from the meta-analyses based on RCTs. It is time to consider whether the resources (human and economic) available can be used better to improve perinatal outcomes than just focusing on the STAN technology. Ellen Blix, Kjetil Gundro Brurberg, Eirik Reierth, Liv Merete Reinar and P al Øian Faculty of Health Sciences, Oslo and Akershus University College of Applied Sciences, Oslo, The Norwegian Knowledge Centre for the Health Services, Oslo, Centre for Evidence Based Practice, Bergen University College, Bergen, Science and Health Library, University Library, UiT The Arctic University of Norway, Tromsø, Department of Obstetrics and Gynecology, University Hospital of North Norway, Tromsø, and Department of Clinical Medicine, UiT The Arctic University of Norway, Tromsø, Norway

Rapid methods including network meta-analysis to produce evidence in clinical decision support: a decision analysis

BackgroundConducting systematic reviews is time-consuming but crucial to construct evidence-based patient decision aids, clinical practice guidelines and decision analyses. New methods might enable developers to produce a knowledge base more rapidly. However, trading off scientific rigour for speed when creating a knowledge base is controversial, and the consequences are insufficiently known. We developed and applied faster methods including systematic reviews and network meta-analyses, assessed their feasibility and compared them to a gold standard approach. We also assessed the feasibility of using decision analysis to perform this comparison.MethodsLong-term treatment in bipolar disorder was our testing field. We developed two new methods: an empirically based, rapid network meta-analysis (NMA) and an expert NMA, and conducted a patient survey. We applied these methods to collect effect estimates for evidence-based treatments on outcomes important to patients. The relative importance of outcomes was obtained from patients using a stated preference method. We used multi-criteria decision analysis to compare a gold standard NMA with the rapid NMA in terms of the ability of the gold standard NMA to change the ranking and expected values of treatments for individual patients.ResultsUsing rapid methods, it was feasible to identify evidence addressing outcomes important to patients. We found that replacing effect estimates from our rapid NMA with estimates from the gold standard NMA resulted in relatively small changes in the ranking and expected value of treatments. The rapid method sufficed to estimate the effects of nine out of ten options. To produce a ranking of treatments accurate for more than 95% of patients, it was necessary to supplement systematic with rapid methods and to use relative importance weights in the analysis. Integrating estimates of the outcome “treatment burden” had a larger impact on rankings than replacing rapid with gold standard methods. Using patients’ importance weights only modestly affected results.ConclusionsThe transfer of knowledge to practice could benefit from faster systematic reviewing methods. The results in this preliminary assessment suggest that an improved rapid NMA approach might replace gold standard NMAs. Decision analysis could be used to compare evidence summarisation methods.

Carbohydrate quantity in the dietary management of type 2 diabetes: A systematic review and meta-analysis

This systematic review and meta‐analysis (registration number: CRD42013005825) compares the effects of low carbohydrate diets (LCDs) on body weight, glycaemic control, lipid profile and blood pressure with the effects of higher carbohydrate diets (HCDs) in adults with type 2 diabetes.

Mortality rates for same-sex married individuals compared with opposite-sex married individuals: potential analytical problems

Mortality rates for same-sex married individuals compared with opposite-sex married individuals – potential analytical problems From Kirsti Malterud,* Norman Anderssen, Kjetil Gundro Brurberg and Guri Rortveit Research Unit for General Practice, Uni Research Health, Bergen, Norway, Department of Global Public Health and Primary Care, University of Bergen, Bergen, Norway, Research Unit for General Practice in Copenhagen, Copenhagen, Denmark, Department of Psychosocial Science, University of Bergen, Bergen, Norway, Primary Care Unit, Norwegian Knowledge Centre for the Health Services, Oslo, Norway and Centre for Evidence Based Practice, Bergen University College, Bergen, Norway