Impact of Misdiagnosis in Case-Control Studies of Myalgic Encephalomyelitis/Chronic Fatigue Syndrome

Abstract

Misdiagnosis of myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) can occur when different case definitions are used by clinicians (relative misdiagnosis) or when failing the genuine diagnosis of another disease (misdiagnosis in a strict sense). This problem translates to a recurrent difficulty in reproducing research findings. To tackle this problem, we simulated data from case-control studies under misdiagnosis in a strict sense. We then estimated the power to detect a genuine association between a potential causal factor and ME/CFS. A minimum power of 80% was obtained for studies with more than 500 individuals per study group. When the simulation study was extended to the situation where the potential causal factor could not be determined perfectly (e.g., seropositive/seronegative in serological association studies), the minimum power of 80% could only be achieved in studies with more than 1000 individuals per group. In conclusion, current ME/CFS studies have suboptimal power under the assumption of misdiagnosis. This power can be improved by increasing the overall sample size using multi-centric studies, reporting the excluded illnesses and their exclusion criteria, or focusing on a homogeneous cohort of ME/CFS patients with a specific pathological mechanism where the chance of misdiagnosis is reduced.

Keywords: ME/CFS; association studies; misclassification; misdiagnosis; simulation; statistical power.

Figure 1

Probabilities of detecting an association (i.e., rejecting $H_0$) as a function of the misdiagnosis rate. Each column represents the values attributed to the risk allele frequency found in matched healthy controls and false positive ME/CFS cases (${\theta }_0\in \{0.05,0.1,0.25,0.5\}$). Each row varies the true odds ratio for the association between risk allele frequency assessed between true positive cases and healthy controls (${\Delta }_T\in \{1.25,1.5,2,3,5,10\}$). Power was estimated for different sample sizes of 100, 250, 500, 1000, 2500, and 5000 ($n_0=n_1$), represented by lines with different colours in each scenario. The upper dashed line indicates the target power of 80% (i.e., $1-\beta =0.80$). The lower dashed line indicates the 5% significance level.

Figure 2

Probabilities of detecting an association (i.e., rejecting $H_0$) as a function of the misdiagnosis rate. Each scenario represents simulated results with a different combination of sensitivity (${\pi }_{se}$) and specificity ${\pi }_{sp}$ for the serological test for columns and rows, respectively. Power was estimated for different sample sizes of 100, 250, 500, 1000, 2500, and 5000 ($n_0=n_1$), represented by lines with different colours in each scenario, with the probability of exposure in healthy controls fixed as ${\theta }_0=0.25$ and true odds ratio ${\Delta }_T=3$. The upper dashed line indicates the target power of 80% (i.e., $1-\beta =0.80$). The lower dashed line indicates the 5% significance level.

Figure 3

The relationship between the misdiagnosis probability (or rate) and the probability of detecting an association (i.e., rejecting the $H_0$) estimated from simulated data from two previously published studies: (A). Data from five different SNPs (genes PTPN22, CTLA4, TNF (TNF1 - rs1799724 and TNF2 - rs1800629), and IRF5); (B). Data of antibody positivity related to six human herpesviruses (CMV, EBV, HSV1 and HSV2, VZV, and HHV6). For each study, risk allele frequencies or the probability of exposure and true odds ratio were determined by Steiner et al. [21] ($n_0=201$; $n_1=305$) and Cliff et al. [22] ($n_0=107$; $n_1=251$; ${\pi }_{se}={\pi }_{sp}=0.975$), with determined values shown in Table 4 and Table 5, respectively. Green lines indicate candidate risk factors where a significant association with the disease was found in the original study. Blue lines show non-significant ME/CFS risk factors. The upper dashed line indicates the target power, where the probability of rejecting the null hypothesis is $1-\beta =0.80$. The lower dashed line indicates the significance level used, $\alpha =0.05$.