| Literature DB >> 32429293 |
Chih-Min Tsai1, Chun-Hung Richard Lin2, Huan Zhang2, I-Min Chiu2,3, Chi-Yung Cheng2,3, Hong-Ren Yu1, Ying-Hsien Huang1.
Abstract
Blood culture is frequently used to detect bacteremia in febrile children. However, a high rate of negative or false-positive blood culture results is common at the pediatric emergency department (PED). The aim of this study was to use machine learning to build a model that could predict bacteremia in febrile children. We conducted a retrospective case-control study of febrile children who presented to the PED from 2008 to 2015. We adopted machine learning methods and cost-sensitive learning to establish a predictive model of bacteremia. We enrolled 16,967 febrile children with blood culture tests during the eight-year study period. Only 146 febrile children had true bacteremia, and more than 99% of febrile children had a contaminant or negative blood culture result. The maximum area under the curve of logistic regression and support vector machines to predict bacteremia were 0.768 and 0.832, respectively. Using the predictive model, we can categorize febrile children by risk value into five classes. Class 5 had the highest probability of having bacteremia, while class 1 had no risk. Obtaining blood cultures in febrile children at the PED rarely identifies a causative pathogen. Prediction models can help physicians determine whether patients have bacteremia and may reduce unnecessary expenses.Entities:
Keywords: bacteremia; children; emergency department; machine learning; predict
Year: 2020 PMID: 32429293 PMCID: PMC7277905 DOI: 10.3390/diagnostics10050307
Source DB: PubMed Journal: Diagnostics (Basel) ISSN: 2075-4418
Distribution of several variables in febrile children with or without bacteremia.
| Variable | Bacteremia | Non-Bacteremia | |
|---|---|---|---|
| Demographics | N = 146 | N = 16,821 | |
| Age | 3.85 ± 4.50 | 5.17 ± 3.97 | <0.001 |
| Gender = Male | 80 (54.79%) | 9308 (55.34%) | 0.896 |
| Gender = Female | 66 (45.21%) | 7513 (44.66%) | |
| Laboratory Values | |||
| WBC (103/μL) | 12.14 ± 6.92 | 11.00 ± 7.33 | 0.062 |
| Neutrophil (%) | 56.35 ± 24.04 | 62.06 ± 19.67 | 0.005 |
| Lymphocyte (%) | 28.93 ± 19.18 | 26.64 ± 16.37 | 0.152 |
| Band (%) | 0.76 ± 1.86 | 0.34 ± 1.54 | 0.007 |
| Monocyte (%) | 7.54 ± 5.38 | 7.80 ± 3.98 | 0.561 |
| Eosinophil (%) | 0.69 ± 1.28 | 0.93 ± 1.57 | 0.067 |
| Basophil (%) | 0.22 ± 0.33 | 0.24 ± 0.36 | 0.501 |
| ANC (103/μL) | 8.57 ± 3.60 | 9.36 ± 2.95 | 0.009 |
| Hemoglobin (g/dL) | 11.68 ± 1.63 | 12.27 ± 1.35 | <0.001 |
| MCV (fL) | 80.51 ± 7.50 | 79.76 ± 6.66 | 0.179 |
| MCH (pg) | 27.19 ± 3.02 | 26.95 ± 2.55 | 0.339 |
| MCHC (g/L) | 33.71 ± 1.38 | 33.75 ± 1.02 | 0.646 |
| Platelet (103/μL) | 279.28 ± 119.06 | 255.50 ± 93.04 | 0.017 |
| AST (U/L) | 42.88 ± 36.44 | 37.45 ± 39.94 | 0.101 |
| ALT (U/L) | 27.75 ± 28.15 | 22.35 ± 34.04 | 0.056 |
| CRP (mg/L) | 53.59 ± 68.24 | 36.46 ± 52.04 | 0.003 |
WBC, white blood cell; ANC, absolute neutrophil count; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration; ALT, alanine transaminase; AST, aspartate transaminase; CRP, C-reactive protein.
Figure 1The distribution of bacteremia at different ages of the study population.
The maximum value, minimum value, mean value, and standard deviation of logistic regression coefficients of characteristic variables.
| Variable | Minimum–Maximum | (Mean ± SD) | ||||||
|---|---|---|---|---|---|---|---|---|
| AGE | −0.0092 | – | 0.0369 | ( | 0.0155 | ± | 0.0084 | ) |
| WBC | −0.1275 | – | −0.0323 | ( | −0.0822 | ± | 0.0176 | ) |
| CRP | 0.0020 | – | 0.0059 | ( | 0.0039 | ± | 0.0008 | ) |
| Hemoglobin | −0.4393 | – | 0.3613 | ( | 0.0119 | ± | 0.1562 | ) |
| MCV | −0.7722 | – | −0.2038 | ( | −0.5010 | ± | 0.1216 | ) |
| MCH | 0.4870 | – | 2.1123 | ( | 1.4471 | ± | 0.3296 | ) |
| MCHC | −1.7403 | – | −0.3591 | ( | −1.0818 | ± | 0.2826 | ) |
| Platelet | −0.0002 | – | 0.0012 | ( | 0.0004 | ± | 0.0003 | ) |
| Lymphocyte | −0.0116 | – | 0.0002 | ( | −0.0053 | ± | 0.0026 | ) |
| AST | −0.0048 | – | 0.0106 | ( | 0.0008 | ± | 0.0026 | ) |
| ALT | −0.0044 | – | 0.0129 | ( | 0.0032 | ± | 0.0033 | ) |
| RBC | −1.6189 | – | 0.6399 | ( | −0.6277 | ± | 0.4279 | ) |
| Band | 0.0003 | – | 0.2040 | ( | 0.0758 | ± | 0.0364 | ) |
| Monocyte | −0.0715 | – | −0.0263 | ( | −0.0491 | ± | 0.0082 | ) |
| Eosinophil | −0.2581 | – | −0.1520 | ( | −0.1999 | ± | 0.0215 | ) |
| Basophil | −0.1549 | – | 0.5387 | ( | 0.1284 | ± | 0.1594 | ) |
| ANC | 0.0731 | – | 0.2211 | ( | 0.1424 | ± | 0.0271 | ) |
| Segment+Band | −0.0282 | – | −0.0152 | ( | −0.0212 | ± | 0.0032 | ) |
WBC, white blood cell; ANC, absolute neutrophil count; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration; ALT, alanine transaminase; AST, aspartate transaminase; CRP, C-reactive protein.
The odds ratio of each variable and its 95%CI after repeating univariate logistic regression 100 times.
| Variable. | OR | OR (95% CI) | Number of | ||||
|---|---|---|---|---|---|---|---|
| AGE | 1.0156 | ( | 0.9632 | – | 1.0685 | ) | 0 |
| WBC | 0.9212 | ( | 0.8239 | – | 1.0184 | ) | 6 |
| CRP | 1.0039 | ( | 1.0008 | – | 1.0069 | ) | 88 |
| Hemoglobin | 1.0243 | ( | 0.4922 | – | 2.3101 | ) | 0 |
| MCV | 0.6104 | ( | 0.3573 | – | 1.0680 | ) | 33 |
| MCH | 4.4778 | ( | 0.8719 | – | 21.6080 | ) | 36 |
| MCHC | 0.3528 | ( | 0.1058 | – | 1.2778 | ) | 26 |
| Platelet | 1.0004 | ( | 0.9985 | – | 1.0024 | ) | 0 |
| Lymphocyte | 0.9947 | ( | 0.9819 | – | 1.0080 | ) | 0 |
| AST | 1.0008 | ( | 0.9933 | – | 1.0071 | ) | 2 |
| ALT | 1.0032 | ( | 0.9944 | – | 1.0112 | ) | 3 |
| RBC | 0.5837 | ( | 0.0609 | – | 4.5206 | ) | 0 |
| Band | 1.0795 | ( | 0.9836 | – | 1.1761 | ) | 35 |
| Monocyte | 0.9522 | ( | 0.9110 | – | 0.9920 | ) | 82 |
| Eosinophil | 0.8190 | ( | 0.6873 | – | 0.9530 | ) | 96 |
| Basophil | 1.1514 | ( | 0.6484 | – | 1.8965 | ) | 0 |
| ANC | 1.1534 | ( | 1.0014 | – | 1.3394 | ) | 43 |
| Segment+Band | 0.9790 | ( | 0.9615 | – | 0.9972 | ) | 82 |
WBC, white blood cell; ANC, absolute neutrophil count; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration; ALT, alanine transaminase; AST, aspartate transaminase; CRP, C-reactive protein.
The maximum value, minimum value, mean value, and standard deviation of multivariate binary logistic regression coefficients of significant risk factors.
| Variable | minimum | – | maximum | ( | Mean | ± | SD | ) |
|---|---|---|---|---|---|---|---|---|
| CRP | 0.0020 | – | 0.0088 | ( | 0.0045 | ± | 0.0010 | ) |
| MONOCYTE | −0.0689 | – | −0.0177 | ( | −0.0434 | ± | 0.0093 | ) |
| EOSINOPHIL | −0.2244 | – | −0.0950 | ( | −0.1789 | ± | 0.0263 | ) |
| Segment+Band | −0.0346 | – | −0.0112 | ( | −0.0244 | ± | 0.0044 | ) |
CRP, C-reactive protein.
Figure 2The relation of FN cost and recall (sensitivity), TNR (specificity), AUC in our predictive model. FN, false negative; TNR, true negative rate; AUC, area under ROC curve.
The maximum value of each performance index in our machine learning model by LR.
| Performance Index | Max Value |
|---|---|
| Sensitivity (recall) | 0.92 (@cost = 12) |
| Specificity (TN rate) | 0.96 (@cos = 7) |
| Positive likelihood ratio | 1.14 (@cost = 9) |
| Negative likelihood ratio | 1.25 (@cost = 11) |
| Positive predictive value | 0.013(@cost = 8) |
| Negative predictive value | 0.993 (@cost = 9) |
| AUC | 0.768 (@cost = 10) |
AUC, area under ROC curve; TN, true negative.
Figure 3The range of risk value and its quartile range.
Figure 4The probability of being bacteremia classified by risk value.
Distribution of several variables in febrile children under the age of 3 years old with or without bacteremia.
| Variable | Bacteremia | Non-Bacteremia | |
|---|---|---|---|
| Demographics | N = 81 | N = 5033 | |
| Age | 0.81 ± 0.78 | 1.12 ± 0.73 | <0.001 |
| Gender = Male | 43 (53.09%) | 2812 (55.87%) | 0.617 |
| Gender = Female | 38 (46.91%) | 2221 (44.13%) | |
| Laboratory Values | |||
| WBC (103/μL) | 12.76 ± 6.40 | 11.40 ± 5.98 | 0.590 |
| Neutrophil (%) | 45.63 ± 21.57 | 48.65 ± 18.75 | 0.151 |
| Lymphocyte (%) | 34.58 ± 17.99 | 37.43 ± 17.19 | 0.140 |
| Band (%) | 0.87 ± 1.88 | 0.39 ±1.64 | 0.026 |
| Monocyte (%) | 8.72 ± 6.38 | 8.91 ± 4.36 | 0.797 |
| Eosinophil (%) | 0.71 ± 1.09 | 1.07 ± 1.53 | 0.004 |
| Basophil (%) | 0.24 ± 0.40 | 0.27 ± 0.35 | 0.591 |
| ANC (103/μL) | 6.98 ± 3.20 | 7.36 ± 2.82 | 0.228 |
| Hemoglobin (g/dL) | 11.20 ± 1.49 | 11.73 ± 1.31 | <0.001 |
| MCV (fL) | 80.43 ± 8.84 | 79.07 ± 7.39 | 0.171 |
| MCH (pg) | 27.09 ± 3.25 | 26.63 ± 2.79 | 0.207 |
| MCHC (g/L) | 33.64 ± 1.05 | 33.64 ± 0.98 | 0.983 |
| Platelet (103/μL) | 296.43 ± 124.97 | 286.64 ± 112.58 | 0.438 |
| AST (U/L) | 46.43 ± 39.59 | 42.80 ± 38.53 | 0.400 |
| ALT (U/L) | 28.43 ± 25.84 | 25.26 ± 30.02 | 0.345 |
| CRP (mg/L) | 60.24 ± 71.39 | 31.47 ± 48.95 | 0.001 |
WBC, white blood cell; ANC, absolute neutrophil count; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration; ALT, alanine transaminase; AST, aspartate transaminase; CRP, C-reactive protein.