Background: We hypothesize that the occurrence of metabolic acidosis correlates with the cumulative rate of gastrointestinal (GI) illness and that incorporating acidosis surveillance would improve models used for the early detection of outbreaks of GI disease.
Methods: We conducted a retrospective cohort study of consecutive patients seen in an urban pediatric tertiary care center from September 1995 to August 2005. All data were analyzed for correlation between acidosis and GI syndrome and for internal periodicities. Four years of data were used to create a model, and the first 100 days of 2004 were used for forecasting. Data collected included visit date, chief complaint (CC), International Classification of Diseases, 9th Revision, diagnoses (Dx), and limited laboratory data. Gastrointestinal syndrome was defined by either CC or Dx. Acidosis was defined as HCO₃ levels 19 mmol/dL or less. Exclusion criteria included hyperglycemia (glucose level >120 mg/dL), glycusoria, or having a test for glycosylated hemoglobin ordered. A simple regression model was used to measure correlation between rates of acidosis and GI_Dx and GI_CC. For acidosis and GI syndrome, we fitted a time series model to the daily data with an auto-regressive integrated moving average (1,1) error term.
Results: During the study period, there were 505,028 emergency department visits. The median age was 5.1 years (interquartile range, 1.6-11.8 years), and 46% of patients were females. Of these, 132,142 had GI_Dx and 136,304 had GI_CC. Blood chemistries were obtained from 91,052 patients (18.1%). Acidosis was detected in 32.4% of patients who had these laboratory tests sent.Periodicities were detected for GI_Dx, GI_CC, acidosis rates affected by day of the week, and seasonality, with no changes in incidence during the years of our study. Acidosis rates highly correlated with rates of GI syndrome on a daily basis (Pearson correlation coefficient, r = 0.66 for GI_Dx and r = 0.68 for GI_CC, P < 0.0001 for both). Having non-diabetic ketoacidosis metabolic acidosis has a 42.2% positive predictive value for GI syndrome by either Dx or CC.Acidosis rates can be forecasted as a stand-alone variable (R² = 0.31, P < 0.001).Adding acidosis rates to time series models for GI_Dx or GI_CC significantly improves forecasting, that is, GI_Dx improved from R² = 0.24 to R² = 0.54, and false alarms rates dropped from 32% to 18%. The GI_CC model improved from R² = 0.32 to R = 0.54, and false alarms rates dropped from 28% to 17%.
Conclusions: Metabolic acidosis rate is a promising data source for real-time disease surveillance in the pediatric population. The rate of metabolic acidosis is highly correlated with the rate of GI syndrome. Adding this variable to currently used models significantly improves forecasting for real-time surveillance.