| Concept | Calculation | Application to child MHPS research in LAMIC |
---|---|---|---|
Area under the curve (AUC) | The probability that the instrument will yield a higher score for a randomly chosen individual with the target condition than for a randomly chosen individual without the condition | Area under the graph with sensitivity on the Y axis by one minus specificity on the X axis | The ideal instrument for screening and/or evaluation of an intervention for children in LAMIC will have a high AUC (close to 1.0). The closer to 0.5 the AUC, the less utility of the screening instrument and the less cost-effectiveness of screening |
Cutoff score | The score on the instrument chosen to differentiate cases from non-cases; may be chosen to maximize specificity, sensitivity, or both | Chosen by researcher based on ROC curve | Based on the type of intervention program, a higher or lower cutoff score could be chosen to prioritize sensitivity or specificity |
Sensitivity | The ability of an instrument, at a selected cutoff score, to identify persons with a target condition. At a sensitivity of 1.0, all persons with the condition are identified, and there are no false negatives |
| Instruments with high sensitivity are ideal to screen children when trying to identify the majority of children in distress needing intervention. At high sensitivity, few children with a condition will be mistakenly deprived of the intervention |
Specificity | The ability of an instrument to include persons who do not have the target condition below the cutoff score. At a specificity of 1.0, no persons without a target condition score above the cutoff |
| Instruments with high specificity minimize the number of children who are incorrectly identified with a high score, but who do not have the target condition. Specificity is a concern when there are negative consequences to being inappropriately included in an intervention, such as stigma or high expense |
Positive predictive value (PPV) | The proportion of persons with scores above cutoff who are correctly classified as having the target condition compared to all persons who score above the cutoff |
| PPV produces more accurate cost estimates of improperly including participants than specificity alone because of accounting for prevalence of a condition in the target population |
Negative predictive value (NPV) | The proportion of persons who score below the selected cutoff who do not have the target condition compared to all persons below the cutoff |
| NPV is used to determine the proportion improperly excluded from an intervention, taking prevalence into account. NPV helps to estimate the cost of not including a proportion of children in an intervention |
Reliability (Cronbach's alpha) | A measures of internal consistency based on the degree of inter-correlation among all items on a scale |
| Reliability is important for newly developed measures or adapted measures in LAMIC to help identify items that may not be culturally or contextually relevant, such as stomachaches in Nepal |