Table 3 Expert panel qualitative assessments of potential prognostic indicators

Fever

Unfavorable

Low specificity for identifying children who will not fail treatment

High feasibility since already implemented and measurement tools simple

Rapid breathing

Minimally favorable

Normal range highly variable

Measurement tools inaccurate

Low sensitivity for hypoxemic children

Low feasibility with respect to sequential monitoring of respiratory rates over time

Performance profile improves with sequential monitoring of respiratory rates over time

Lower chest wall indrawing

Favorable

Low specificity for identifying children who will not fail treatment

Subjective sign with variable inter-provider agreement levels

Abnormal oxygen saturation

Highly favorable

High specificity for identifying children who will not fail treatment

Objective measurement but inter-provider agreement levels unknown

Later indicator decreases sensitivity for identifying children who will fail treatment at the community-level

Robust, precise, low-cost point-of-care instrument needed

Likely costly to implement and sustain

High blood lactate levels

Unfavorable

Low specificity for identifying children who will not fail treatment

Point-of-care tool appropriate for community use not available, likely costly

Very late indicator decreases sensitivity for identifying children who will fail treatment at the community-level

Moderate malnutrition

Highly favorable

Mid-range sensitivity and specificity which likely improves greatly in combination with other indicators

High feasibility since measurement tools simple and accurate

High positive predictive value

HIV-affected

Minimally favorable

Geographically limited in relevancy (e.g. primarily in eastern and southern Africa)

Low feasibility since difficult to obtain HIV status at community-level (e.g. stigma)

High positive predictive value