Table 5.
Hyperparameter sensitivity analysis—estimated risk differences for single risk factors.
| n | Risk difference | |||
|---|---|---|---|---|
| Average | Minimum | Maximum | ||
| Full population | 88,818 | 3.31 | 3.29 | 3.32 |
| Female | 57,087 | 3.94 | 3.93 | 3.95 |
| Male | 31,731 | 2.18 | 2.15 | 2.20 |
| Older than 50 years | 40,439 | 4.27 | 4.24 | 4.29 |
| Fibromyalgia | 803 | 10.36 | 10.18 | 10.54 |
| COPD or other lung disease | 1197 | 6.89 | 6.79 | 7.04 |
| Diabetes | 2980 | 6.35 | 6.27 | 6.49 |
| Post-traumatic stress disorder | 1757 | 5.79 | 5.69 | 5.87 |
| High BMI | 14,700 | 5.66 | 5.60 | 5.69 |
| Depression | 10,715 | 5.65 | 5.59 | 5.71 |
| Chronic asthma | 6335 | 5.40 | 5.27 | 5.46 |
| High blood pressure | 9817 | 5.00 | 4.98 | 5.04 |
| Chronic or frequent headaches | 3593 | 4.99 | 4.86 | 5.05 |
| Chronic fatigue syndrome | 1345 | 3.24 | 3.10 | 3.32 |
| Anxiety | 7425 | 3.40 | 3.34 | 3.47 |
The average, minimum, and maximum risk difference for the full population and different single risk factors obtained from a sensitivity analysis repeating the main analysis with different combinations of the hyperparameters sample fraction, mtry, minimum node size, and honesty fraction used by the causal_forest function from the grf R package. Three values of each hyperparameter was used, for a total of 81 configurations tested. Results were obtained using CF and AIPW for substantial full-time sick leave taken 4 weeks to 9 months after the test date between SARS-CoV-2 test-positives and test-negatives for the total study population and possible PCC risk groups.