Background
Methods
Subject cohorts
Ethics statement
Cytokine profiles
Statistical analysis
Results
Differences in cytokine expression between healthy and ME/CFS populations
Validating a prior classification model
Classification based on | Stepwise model of IL-1α, 6,8,13,23 | Consensus model of IL-6, 8, and 23 | Minimal model of IL-6 and 8 |
---|---|---|---|
(a) Training set: Age < =18 years | |||
Correct rate | 0.93 | 0.86 | 0.79 |
Error rate | 0.07 | 0.14 | 0.21 |
Sensitivity | 0.94 | 0.89 | 0.78 |
Specificity | 0.92 | 0.83 | 0.79 |
Positive predictive value | 0.89 | 0.80 | 0.74 |
Negative predictive value | 0.96 | 0.91 | 0.83 |
(b) Test set: 18 < Age < = 50 years | |||
Correct rate | 0.49 | 0.26 | 0.35 |
Error rate | 0.51 | 0.74 | 0.65 |
Sensitivity | 0.09 | 0.23 | 0.09 |
Specificity | 0.90 | 0.29 | 0.62 |
Positive predictive value | 0.50 | 0.25 | 0.20 |
Negative predictive value | 0.49 | 0.26 | 0.39 |
(c) Test set: Age > 50 years a
| |||
Correct rate | 0.46 | 0.29 | 0.39 |
Error rate | 0.54 | 0.71 | 0.61 |
Sensitivity | 0.00 | 0.32 | 0.14 |
Specificity | 0.93 | 0.25 | 0.64 |
Positive predictive value | 0.00 | 0.30 | 0.29 |
Negative predictive value | 0.48 | 0.27 | 0.43 |
Selecting cytokines broadly conserved across age and BMI
(a) HC all age groups | (b) HC groups age ≤50 years only | |||
---|---|---|---|---|
p ANOVA | p Kruskalwallis | p ANOVA | p Kruskalwallis | |
BMI | 0.00 | 0.01 | 0.00 | 0.00 |
IL-1α |
0.23
| 0.04 |
0.12
|
0.12
|
IL-1β | 0.00 | 0.00 | 0.00 | 0.00 |
IL-2 |
0.27
| 0.01 |
0.32
| 0.02 |
IL-4 | 0.00 | 0.00 |
0.18
| 0.01 |
IL-5 | 0.00 | 0.00 | 0.00 | 0.00 |
IL-6 |
0.99
|
0.89
|
0.91
|
0.95
|
IL-8 |
0.08
| 0.01 |
0.27
|
0.06
|
IL-10 | 0.00 | 0.00 | 0.00 | 0.00 |
IL-12p70 | 0.00 | 0.00 | 0.02 | 0.01 |
IL-13 | 0.00 | 0.00 | 0.00 | 0.00 |
IL-15 |
0.22
| 0.04 |
0.98
|
0.41
|
IL-17 | 0.00 | 0.00 | 0.00 | 0.01 |
IL-23 | 0.00 | 0.00 | 0.00 | 0.00 |
IFNγ | 0.02 | 0.02 | 0.00 | 0.01 |
TNFα |
0.89
|
0.33
|
0.82
|
0.17
|
LTα |
0.11
|
0.75
|
0.09
|
0.67
|
Subject sub-group | Classification model | ||
---|---|---|---|
Training selection | Training selection | ||
(a) Training set: Age < =18 years | IL-1α, 6, 8 | IL-1α, 6, 8 | |
Correct rate | 0.88 | 0.88 | |
Error rate | 0.12 | 0.12 | |
Sensitivity | 0.89 | 0.89 | |
Specificity | 0.88 | 0.88 | |
Positive predictive value | 0.84 | 0.84 | |
Negative predictive value | 0.91 | 0.91 | |
Training selection | Test set | Parameters tuned | |
(b) Training set: 18 < Age < = 50 years | IL-1α, 15 | IL-1α, 6, 8 | IL-1α, 6, 8 |
Correct rate | 0.72 | 0.37 | 0.77 |
Error rate | 0.28 | 0.63 | 0.23 |
Sensitivity | 0.73 | 0.09 | 0.77 |
Specificity | 0.71 | 0.67 | 0.76 |
Positive predictive value | 0.73 | 0.22 | 0.77 |
Negative predictive value | 0.71 | 0.41 | 0.76 |
(c) Training set: Age > 50 years a
| IL-6, 8, 15, TNFα | IL-1α, 6, 8 | IL-1α, 6, 8 |
Correct rate | 0.84 | 0.48 | 0.75 |
Error rate | 0.16 | 0.52 | 0.25 |
Sensitivity | 0.86 | 0.18 | 0.75 |
Specificity | 0.82 | 0.79 | 0.75 |
Positive predictive value | 0.83 | 0.45 | 0.75 |
Negative predictive value | 0.85 | 0.49 | 0.75 |
A classification model corrected for duration of illness
Classification based on IL-1α, 6 and 8 | Median (MADM Std Err a) | Mean (Std Err) |
---|---|---|
(a) 50 random subsets: Age < =18 years | ||
Correct rate | 0.90 (0.01) | 0.89 (0.06) |
Sensitivity | 0.90 (0.00) | 0.92 (0.06) |
Specificity | 0.85 (0.01) | 0.86 (0.08) |
Positive predictive value | 0.87 (0.01) | 0.87 (0.07) |
Negative predictive value | 0.90 (0.00) | 0.92 (0.06) |
(b) 50 random subsets: 18 < Age < = 50 years | ||
Correct rate | 0.75 (0.01) | 0.76 (0.08) |
Sensitivity | 0.80 (0.02) | 0.78 (0.09) |
Specificity | 0.70 (0.02) | 0.73 (0.11) |
Positive predictive value | 0.73 (0.01) | 0.75 (0.10) |
Negative predictive value | 0.78 (0.01) | 0.77 (0.08) |
(c) 50 random subsets: Age > 50 years b
| ||
Correct rate | 0.80 (0.01) | 0.79 (0.08) |
Sensitivity | 0.80 (0.02) | 0.75 (0.12) |
Specificity | 0.80 (0.01) | 0.83 (0.08) |
Positive predictive value | 0.80 (0.01) | 0.82 (0.08) |
Negative predictive value | 0.80 (0.02) | 0.77 (0.09) |
(a) Median (MADM-based Std Err a) | Age < =18 years | 18 < Age < = 50 years | Age > 50 years | |
---|---|---|---|---|
Discriminant coefficients (n = 50) | ||||
IL-1α | α1
| 1.93 (0.11) | 0.51 (0.06) | 0.14 (0.10) |
IL-6 | α2
| −2.08 (0.17) | 0.91 (0.12) | 1.28 (0.07) |
IL-8 | α3
| 2.89 (0.18) | −0.60 (0.06) | −1.36 (0.12) |
Constant term | α0
| −1.86 (0.09) | −0.39 (0.05) | −0.49 (0.05) |
Duration of illness (years) | 2.00 (0.00) | 6.50 (1.74) | 8.50 (1.26) | |
(b) Mean (Std Err) | ||||
Discriminant coefficients (n = 50) | ||||
IL-1α | α1
| 2.41 (0.23) | 0.53 (0.10) | 0.14 (0.13) |
IL-6 | α2
| −2.36 (0.19) | 1.09 (0.13) | 1.51 (0.13) |
IL-8 | α3
| 3.53 (0.27) | −0.92 (0.14) | −1.52 (0.19) |
Constant term | α0
| −2.07 (0.15) | −0.48 (0.06) | −0.61 (0.09) |
Duration of illness (years) | 2.00 (0.00) | 7.09 (1.28) | 10.64 (1.46) |
Regression Model | Intercept β0
| β1 (years ill) | β2 (years ill 2) | R2
| F | p | |
---|---|---|---|---|---|---|---|
IL-1a coefficient | α1
| 3.42 (2.80, 4.04) | −0.75 (−0.97, −0.52) | 0.04 (0.02, 0.06) | 0.347 | 38.980 | 0.000 |
IL-6 coefficient | α2
| −4.11 (−4.80, −3.43) | 1.31 (1.06, 1.56) | −0.07 (−0.09, −0.05) | 0.573 | 98.510 | 0.000 |
IL-8 coefficient | α3
| 5.84 (4.95, 6.73) | −1.71 (−2.04, −1.38) | 0.09 (0.07, 0.12) | 0.573 | 98.504 | 0.000 |
Constant term | α0
| −2.56 (−3.00, −2.12) | 0.45 (0.29, 0.62) | −0.02 (−0.04, −0.01) | 0.310 | 32.986 | 0.000 |