Application of two-phase statistical process control to detect glycemic instability in critically ill patients

To apply a two-phase statistical process control (SPC) framework to monitor ICU Day-2 glycemic performance, detect SPC-defined glycemic instability, and identify clinical predictors of out-of-control Day-2 glycemic patterns in a tertiary hospital ICU in Egypt's Delta region.

A prospective observational cohort study was conducted in an adult mixed medical-surgical ICU. Phase I data included 300 patients, each with four glucose measurements on ICU Day-2, and were used to estimate empirical x̄-chart control limits. Phase II surveillance applied the fixed Phase I limits to 1,000 additional patients, each with four ICU Day-2 glucose measurements. Nelson Rules 1 and 2 were used as pre-specified operational signal-detection criteria to identify extreme excursions and sustained unit-level process shifts. Multivariable logistic regression was used to explore clinical factors associated with SPC-defined out-of-control Day-2 glycemic status.

In Phase II, 505/1,000 (50.5%) patient subgroups signaled "out-of-control" by the union of Nelson rules. Rule-1 limit exceedances occurred in 150 subgroups, while Rule-2 shift signaling affected 413 subgroups (35 distinct run events). In adjusted analyses, higher illness severity (APACHE II), sepsis, older age, higher baseline HbA1c, vasopressor use, corticosteroid use, and comorbidity burden (CKD and/or liver disease) increased the odds of "out-of-control" status; higher caloric intake was associated with lower odds. BMI, sex, and surgical admission type were not significant predictors.

Observational design precludes causal inference and assessment of real-time SPC impact. Excluding DKA/HHS and stays <48 h may limit generalizability. Subgrouping (n = 4) favors shift detection but may mask very short-term variability. Single-center results may not generalize to ICUs with different nutrition/sedation and glycemic management practices.

A combined SPC-and-regression framework can provide early operational surveillance of ICU Day-2 glycemic instability and may support targeted, data-driven quality-improvement interventions. However, prospective interventional studies are required to determine whether SPC-guided glycemic management improves clinical outcomes.

This study combines Phase-I-derived SPC limits with Phase II surveillance and multivariable risk modeling to measure and explain Day-2 glycemic special-cause variation in a resource-relevant tertiary ICU environment.