Classification of a Cohort of Acute Pancreatitis Patients
Classification of a Cohort of Acute Pancreatitis Patients
The Severity of Acute Pancreatitis Study (SAPS) was conducted in three 1-year phases between 2003 and 2010 at the University of Pittsburgh Medical Center in Pittsburgh, PA. In each 1-year periods approximately 60–100 consecutive patients were prospectively enrolled. Each patient signed written informed consent before entering the study. Ethical approval was obtained from the Institutional Review Board (IRB protocol ID PRO08010374).
Patients were enrolled in the study at the time of admission and prospectively followed until hospital discharge and during subsequent admissions. Patients transferred from other hospitals were included in the study; outside hospital medical records were obtained at the time of enrollment, reviewed, and pertinent data elements (i.e., admission vital signs and laboratory values) were abstracted.
The diagnosis of AP was based on the presence of at least two of the following three criteria: (i) abdominal pain suggestive of AP, (ii) elevation in serum amylase and/or lipase >3 times the upper limit of normal, and (iii) computed tomography (CT) findings characteristic of AP. Presence of OF was assessed at admission and every 24 h during the hospitalization. More specifically, transient OF was defined as OF involving the respiratory, cardiovascular, or renal systems lasting <48 h, whereas persistent OF was defined as OF in any of the three organ systems lasting ≥48 h as previously described. The modified Marshall scoring system and sepsis-related OF score were not calculated.
Data pertaining to imaging (contrast enhanced CT scan) were ascertained with regards to local complications as previously described. In addition, CT scans of patients with OF or local complications were recently reviewed by an abdominal radiologist using the Revised Atlanta Classification 2012 definitions (including acute peripancreatic fluid collection, acute necrotic collection, (peri)Pnec, and Inf(peri)Pnec (presence of extraluminal gas). The diagnosis of Inf PNec was confirmed by fine-needle aspiration of necrotic tissue or during minimally invasive or surgical intervention.
The three AP severity classification systems: Atlanta 1992, Atlanta 2012, and DBC were applied to this prospectively enrolled cohort of patients (Table 1). For each patient, the peak severity category during the entire follow-up period (12 months) was selected for each classification system given that interventions were frequently performed following the index admission.
Primary outcomes included mortality, intensive care unit (ICU) admission, need for surgical or minimally invasive interventions, ICU length of stay (LOS), and overall hospital LOS. These outcomes have been previously reported in validation studies for classification of AP severity. Hospital LOS reflected the index hospitalization and included the days in the outside hospital for transferred patients. Interventions were defined as surgical (surgical cystgastrostomy, open and laparoscopic pancreatic debridement, and exploratory laparotomy for abdominal compartment syndrome), endoscopic (cystgastrostomy and direct endoscopic necrosectomy), and interventional (percutaneous drain placement).
Categorical variables were described using frequencies and percentages. Continuous variables were summarized using mean and s.d. or median and interquartile range depending on the distribution. All five clinical outcomes were stratified by severity categories for each classification system.
The overall association between clinical outcomes and severity categories was tested using Cochrane–Armitage trend test for binary outcomes (mortality, ICU admission, and need for interventions) and Cuzick's trend test for continuous outcomes (ICU and hospital LOS). Cochrane–Armitage and Cuzick's trend tests are modified versions of Pearson's χ and Wilcoxon tests used to incorporate a suspected ordering in the effects of one of the co-variates (in our case, the severity categories of the classification systems).
Pairwise testing between severity grades within a classification system was performed using Fisher's exact and Kruskal–Wallis tests for binary and continuous outcomes respectively. The Kruskal–Wallis test compares median values between groups and is a nonparametric alternative of the analysis of variance for skewed distributions. Pairwise comparison was performed in a step-wise manner; that is, mild vs. moderate and moderate vs. severe categories for the Atlanta 2012 classification. The Sidak's adjustment, a more robust alternative to the Bonferronni correction method for multiple testing, was used to control for the type I error rate from multiple testing in pairwise comparisons.
For binary outcomes, the area under the ROC curve (AUC) was used to describe the predictive accuracy of a classification system, whereas the Somer's D was reported for continuous outcomes. AUC is a rescaled version of Somer's D for binary outcomes. Pairwise comparison of AUC and Somer's D was performed with Sidak's adjustment to control for type I error. Significance level was chosen at 0.05. The entire analysis was performed using Stata Statistical Software Release 12 (StataCorp LP, College Station, TX).
Methods
Patient Population and Data Collection
The Severity of Acute Pancreatitis Study (SAPS) was conducted in three 1-year phases between 2003 and 2010 at the University of Pittsburgh Medical Center in Pittsburgh, PA. In each 1-year periods approximately 60–100 consecutive patients were prospectively enrolled. Each patient signed written informed consent before entering the study. Ethical approval was obtained from the Institutional Review Board (IRB protocol ID PRO08010374).
Patients were enrolled in the study at the time of admission and prospectively followed until hospital discharge and during subsequent admissions. Patients transferred from other hospitals were included in the study; outside hospital medical records were obtained at the time of enrollment, reviewed, and pertinent data elements (i.e., admission vital signs and laboratory values) were abstracted.
The diagnosis of AP was based on the presence of at least two of the following three criteria: (i) abdominal pain suggestive of AP, (ii) elevation in serum amylase and/or lipase >3 times the upper limit of normal, and (iii) computed tomography (CT) findings characteristic of AP. Presence of OF was assessed at admission and every 24 h during the hospitalization. More specifically, transient OF was defined as OF involving the respiratory, cardiovascular, or renal systems lasting <48 h, whereas persistent OF was defined as OF in any of the three organ systems lasting ≥48 h as previously described. The modified Marshall scoring system and sepsis-related OF score were not calculated.
Data pertaining to imaging (contrast enhanced CT scan) were ascertained with regards to local complications as previously described. In addition, CT scans of patients with OF or local complications were recently reviewed by an abdominal radiologist using the Revised Atlanta Classification 2012 definitions (including acute peripancreatic fluid collection, acute necrotic collection, (peri)Pnec, and Inf(peri)Pnec (presence of extraluminal gas). The diagnosis of Inf PNec was confirmed by fine-needle aspiration of necrotic tissue or during minimally invasive or surgical intervention.
The three AP severity classification systems: Atlanta 1992, Atlanta 2012, and DBC were applied to this prospectively enrolled cohort of patients (Table 1). For each patient, the peak severity category during the entire follow-up period (12 months) was selected for each classification system given that interventions were frequently performed following the index admission.
Outcomes
Primary outcomes included mortality, intensive care unit (ICU) admission, need for surgical or minimally invasive interventions, ICU length of stay (LOS), and overall hospital LOS. These outcomes have been previously reported in validation studies for classification of AP severity. Hospital LOS reflected the index hospitalization and included the days in the outside hospital for transferred patients. Interventions were defined as surgical (surgical cystgastrostomy, open and laparoscopic pancreatic debridement, and exploratory laparotomy for abdominal compartment syndrome), endoscopic (cystgastrostomy and direct endoscopic necrosectomy), and interventional (percutaneous drain placement).
Statistical Analysis
Categorical variables were described using frequencies and percentages. Continuous variables were summarized using mean and s.d. or median and interquartile range depending on the distribution. All five clinical outcomes were stratified by severity categories for each classification system.
The overall association between clinical outcomes and severity categories was tested using Cochrane–Armitage trend test for binary outcomes (mortality, ICU admission, and need for interventions) and Cuzick's trend test for continuous outcomes (ICU and hospital LOS). Cochrane–Armitage and Cuzick's trend tests are modified versions of Pearson's χ and Wilcoxon tests used to incorporate a suspected ordering in the effects of one of the co-variates (in our case, the severity categories of the classification systems).
Pairwise testing between severity grades within a classification system was performed using Fisher's exact and Kruskal–Wallis tests for binary and continuous outcomes respectively. The Kruskal–Wallis test compares median values between groups and is a nonparametric alternative of the analysis of variance for skewed distributions. Pairwise comparison was performed in a step-wise manner; that is, mild vs. moderate and moderate vs. severe categories for the Atlanta 2012 classification. The Sidak's adjustment, a more robust alternative to the Bonferronni correction method for multiple testing, was used to control for the type I error rate from multiple testing in pairwise comparisons.
For binary outcomes, the area under the ROC curve (AUC) was used to describe the predictive accuracy of a classification system, whereas the Somer's D was reported for continuous outcomes. AUC is a rescaled version of Somer's D for binary outcomes. Pairwise comparison of AUC and Somer's D was performed with Sidak's adjustment to control for type I error. Significance level was chosen at 0.05. The entire analysis was performed using Stata Statistical Software Release 12 (StataCorp LP, College Station, TX).