A Method Using Longitudinal Laboratory Data to Predict Future Intestinal Complication in Patients with Crohn's Disease

Abstract

Background: Stenosis, fistulization, and perforation of the bowel are severe outcomes which can occur in patients with Crohn's disease. Accurate prediction of these events may enable clinicians to alter treatment strategies and avoid these outcomes.

Aims: To study the correlation between longitudinal laboratory testing and subsequent intestinal complications in patients with Crohn's disease.

Methods: An observational cohort of patients with Crohn's disease at a single center were analyzed between 01/01/1994 and 06/30/2016. A complication was defined as the development of an intestinal fistula, stenosis, or perforation. Exploratory analysis using Cox regression was performed to select the best statistical method to represent longitudinal laboratory data. Cox regression was used to identify laboratory variables independently associated with the development of a subsequent complication. A clinical scoring tool was designed.

Results: In 246 patients observed over a median of 5.72 years, 134 complications occurred. Minimum or maximum value in a preceding window period of one year was most strongly associated with subsequent complication. A Longitudinal Laboratory score of ≥ 2 (maximum albumin level < 39 g/L = 1, maximum mean cell volume < 88 fL = 1, minimum platelet count > 355 × 10⁹/L = 1, minimum C reactive protein > 5 mg/L = 1) was 62% sensitive and 91% specific in identifying patients who develop a subsequent complication.

Conclusion: A consistent reduction in serum albumin and mean cell volume, and a consistent increase in platelet count and C reactive protein were associated with a subsequent complication in patients with Crohn's disease. Longitudinal laboratory tests may be used as described in this paper to provide a rational for earlier escalation of therapy.

Keywords: Biomarkers; Crohn’s disease; Fistula; Inflammatory bowel diseases; Perforation; Stenosis.

Fig. 1

Z value for univariate cox proportional hazard analyses of association between complication and laboratory variable (albumin, CRP, MCV, and platelet count), using varying statistical methods to represent the laboratory variable. The event horizon decreases from 720 to 0 days from left to right in each graph. For variables that tend to rise with illness (CRP, platelet count), minimum value during window period was the representative statistic most strongly associated with complication. For variables that tend to fall with illness (albumin, hemoglobin, mean cell volume), maximum value was most strongly associated with complication

Fig. 2

Z value for univariate cox proportional hazard analyses of association between complication, and the selected statistical method to represent laboratory variables (albumin, CRP, MCV, and platelet count), with a varying window period. The event horizon decreases from 720 to 0 days from left to right. 365 days is the window period that is most consistently correlated with complication

Fig. 3

Z scores for univariate cox proportional hazard analyses of each pathology test, each modeled independently using minimum (white cell count, neutrophil count, platelet count, ALT, CRP, ESR, fecal calprotectin) or maximum (albumin, hemoglobin, ferritin, MCV) of values from a 365-day window period prior to each time point. The event horizon decreases from 720 to 0 days from left to right. Values above the dashed line survive Bonferroni correction

Fig. 4

Z score for univariate cox regression of laboratory variables categorized above or below a cut-off. The cut-off used varies along the x axis, and the vertical dotted line represents the cut-off value which gives the strongest association with complication

Fig. 5

Receiver Operating Characteristic (ROC) curve for the Longitudinal Laboratory Scoring Tool in predicting a subsequent complication