Introduction: Laboratory test overuse is associated with harm in critically ill children, and quantifying the informational value of tests may help reduce unnecessary testing. A test’s entropy reflects its inherent uncertainty and can be used to calculate the novel information (NI) in repeated testing. NI is the amount of information in the conditional distribution of test values given prior results; an NI value of X bits corresponds to 2X equally likely outcomes (e. g. , an 8-sided die has 2³ outcomes = 3 bits of information). Low NI indicates high predictability given prior results. We hypothesized that some repeated laboratory tests have low NI, and that this varies by age. Methods: We retrospectively analyzed the PICU Data Collaborative 8-site dataset, including ICU encounters with at least one complete blood count (CBC) or basic metabolic panel (BMP) daily on ICU Days 1-3. We used the median value per day for each component. After outlier removal, component values were discretized into 20 equal-width bins (maximum NI: 4. 32 bits). We calculated NI on Day 2 given Day 1, and on Day 3 given Day 2. Patients were grouped as infants (< 1 year old) or older children (≥1 year old). Differences between infants and older children were calculated within each site, and these paired site-level differences were analyzed using the Wilcoxon signed-rank test. Results: There were 23, 102 CBC sets (26. 6% were from infants) and 29, 532 BMP sets (32. 9% infants). Creatinine had the lowest NI (Day 2/3 values: 1. 25/1. 07 bits), and potassium had the highest (3. 49/3. 42 bits). Infants had lower NI than older children for creatinine (Day 2: mean difference -0. 25 bits, 95% CI -0. 43, -0. 07, p=0. 023; Day 3: -0. 22 bits -0. 36, -0. 08, p=0. 007) and for bicarbonate on Day 2 (-0. 16 bits -0. 29, -0. 04, p=0. 023). Older children had lower NI for platelets (Day 2: -0. 22 bits -0. 32, -0. 12, p=0. 007; Day 3: -0. 22 bits -0. 29, -0. 15, p=0. 007). No other components varied significantly by age. Conclusions: Repeated testing of creatinine and bicarbonate in infants yields less NI compared to older children; the opposite holds true for platelets. These findings support targeted reductions in laboratory testing. Extending these methods to include specific patient factors may help guide laboratory testing decision-making.
Shah et al. (Sun,) studied this question.