Gastroenterology

Development of a prediction model for clinically important outcomes of acute diverticulitis

a b s t r a c t

Objective: acute diverticulitis is a common disease with various outcomes. When AD is diagnosed in the emergency department (ED), the ED clinician must determine the patient’s treatment strategy whether the pa- tient can be discharged, needs to be admitted to the general ward, ICU, or needs Surgical consultation. This study aimed to identify potential risk factors for clinically important outcomes (CIOs) and to develop a prediction model for CIOs in AD to aid clinical decision making in the ED.

Methods: Retrospective data from between 2013 and 2017 in an ED in an urban setting were reviewed for adult AD. Potential risk factors were age, sex, past medical history, symptoms, physical exams, laboratory results, and Imaging results. A CIO was defined as a case with one of the following outcomes: hospital death, ICU admission, surgery or invasive intervention, and admission for 7 or more days. The prediction model for CIOs was developed using potential risk factors. Model discrimination and calibration were assessed using the area under the curve (AUC) and 95% confidence intervals (CIs) and the Hosmer-Lemeshow (HL) test, respectively. Model validation was conducted using 500 random bootstrap samples.

Results: Of the final 337 AD patients, 63 patients had CIOs. Six potential factors (age, abdominal pain (>= 3 days), anorexia, rebound tenderness, white blood cell count (> 15,000/ul), C-reactive protein (> 10 mg/dL), and CT findings of a complication) were used for the final model. The AUC (95% CI) for CIOs was 0.875 (0.826-0.923), and ?2 was 2.969 (p-value = 0.936) with the HL test. Validation using bootstrap samples resulted in an optimism-corrected AUC of 0.858 (0.856-0.861).

Conclusion: A prediction model for clinically important outcomes of AD visiting a single ED showed good discrim- ination and calibration power with an acceptable range.

(C) 2021

  1. Introduction

Diverticulitis, a disease defined as inflammation of the diverticulum or diverticula, is common, and the incidence rate has been reported to be up to 115/100,000 person-years [1,2]. The incidence rate of hospital admission for diverticulitis has been reported to be 62/100,000 person-years in Caucasians and 10/100,00 person-years for Asians [3]. In the Caucasian population diverticulitis of the right side of the colon is rare and has been reported to consist of under 10% of patients with di- verticulitis [4,5]. Whereas in Asian populations, over 85% of diverticulitis has been reported to occur in the right-colon [6]. Diverticulitis in the

* Corresponding author at: Department of Emergency Medicine, Seoul National University Hospital, 101 Daehak-Ro, Jongno-Gu, Seoul 03080, Republic of Korea.

E-mail addresses: [email protected] (S.G.W. Lee), [email protected] (S.D. Shin), [email protected] (H.J. Lee), [email protected] (G.J. Suh), [email protected] (D.J. Park).

right-colon has been suggested to be less severe than that of the left- colon [7,8]. Law et al. reported that postoperative mortality after emer- gency surgery for diverticulitis was 13% in the left-colon diverticulitis group and 0% in the right-colon diverticulitis group [7].

The most common clinical presentation of acute diverticulitis (AD) is abdominal pain, followed by nausea, diarrhea, vomiting, constipation, and fever, while hematochezia is rare [9,10]. Computed tomography (CT) has been conventionally recommended for initial radiological ex- amination with sensitivity and specificity of up to 95% and 100% [2,11,12]. The treatment of diverticulitis differs according to the Severity of disease, and treatment strategies have continuously evolved from ag- gressive surgical management to more conservative management. Re- cent guidelines state that oral antibiotics and outpatient follow-up are effective and safe for patients with uncomplicated diverticulitis [1,2], while a few studies have suggested that follow-up without antibiotics might also be safe [12-14]. The ability to differentiate patients with non-severe disease from those with severe disease in the emergency

https://doi.org/10.1016/j.ajem.2021.06.071

0735-6757/(C) 2021

department (ED) is important to determine who should be discharged and who should be admitted. The current evidence suggests that disease severity evaluated by CT findings should be used for decision making [15,16].

The severity of diverticulitis has been conventionally evaluated using CT findings based on the Hinchey classification, which categorizes diverticulitis into four groups according to the existence of pericolic phlegmon, inflammation, fluid collection, abscess, and peritonitis [2]. In addition, the Ambrosetti CT staging for diverticulitis categorizes mod- erate diverticulitis and severe diverticulitis based on CT findings [11]. Previous studies have evaluated risk factors for severe diverticulitis with or without CT findings. Patient history, such as age, sex, history of previous diverticulitis, comorbidities, and immunosuppression; clin- ical signs, such as nausea, vomiting, constipation, fever, generalized ab- dominal pain, rebound tenderness, and guarding; and laboratory results of high white blood cell count , high C-reactive protein (CRP) count, and complications on CT [17-20] have been reported to be risk factors for severe diverticulitis. However, few studies have reported on the clinical decision rules for predicting clinically important outcomes after AD, in addition to severity classifications using clinical, laboratory, and imaging findings.

The ED is a common point of diagnosis and entry to the hospital for patients with AD [21]. When AD is diagnosed in ED, the ED clinician must determine the patient’s treatment strategy whether the patient can be discharged, needs to be admitted to the general ward, ICU, or needs surgical consultation. The clinician’s ability to adequately predict patient outcomes is critically important for adequate patient disposition in the ED. The goal of this study was to develop and validate a prediction model for clinically important outcomes in patients diagnosed with AD in the ED that can be applied in the ED.

  1. Methods
    1. Study design and setting

This study was a retrospective, single-center, observational study. An electronic medical records (EMR) review of patients diagnosed with AD in an urban teaching hospital ED located in Seoul between Jan- uary 2013 and July 2017 was performed. The hospital has approxi- mately 400 doctors and 761 beds, including 53 ICU beds. The annual numbers of outpatient visits and admissions are approximately 850,000 and 240,000, respectively. The study hospital serves an Asian population where right-colon diverticulitis is dominant. Previous stud- ies suggest that patients with right-colon diverticulitis have lower mor- tality, shorter hospital stay, and need surgery less frequently when compared to left-colon diverticulitis [7,22].

The EMR system in the hospital enables physicians to record patient histories, display test results, write prescriptions, and enter orders. In addition, a picture archiving and communication system (PACS) is used to store and view images generated from radiology exams, such as CT or X-rays. The Radiology reports can be viewed both in the EMR system and in PACS.

The annual ED visits numbered 58,000 in 2019. The ED with 43 beds is staffed by professors, residents, interns, nurses, nurse assistants, and emergency medical technicians. Patients who walk in or use emergency medical services in the ED are initially assessed by a certified triage nurse after patient registration. The Korean Triage and Acuity Scales (KTAS) program is used for ED triage by the Triage nurses. After triage is performed, the patient is assigned to an emergency medicine resi- dent, who orders laboratory exams and imaging exams by an ED resi- dent protocol after clinical evaluation. The assigned resident uses a review of systems and physical examination template during initial examination.

The process from triage to Order entry is performed regularly within 20 min. Routine laboratory exams performed in the ED are as follows: complete blood cell count (white blood cell count, hemoglobin),

electrolyte panel (sodium, potassium, chloride), admission panel (blood urea nitrogen, creatinine, glucose, total bilirubin, aspartate ami- notransferase, alanine aminotransferase, albumin, alkaline phospha- tase), coagulation panel (activated partial thromboplastin time, prothrombin time, international normalized ratio), and CRP. Whole pro- cesses of ED care are supervised by an emergency medicine (EM) physician.

When the results of the exams are reported, the assigned resident decides the patient’s disposition: admission, transfer, or discharge. If ad- mission is needed, the emergency medicine resident consults an appro- priate specialist after the EM physicians’ confirmation. During the study period, no standard protocol for AD at the ED was used. All of the pa- tients were treated by the physician’s preference rather than the depart- ment protocol.

    1. Study population

Patients 18 years of age or older who were diagnosed with AD dur- ing the study period in the ED were included. Patients who were diag- nosed with AD in the outpatient clinic and transferred to the ED were excluded. Patients with AD other than that of the colon, such as of the Small intestine, bladder and esophagus, were excluded. Because the di- agnosis of diverticulitis was based on CT evaluation, patients without abdominal CT scans were also excluded. Patients without results of any of the routinely ordered laboratory tests (complete blood cell count, electrolyte panel, admission panel, coagulation panel, and CRP) were excluded. Patients without vital signs checked upon ED arrival were excluded. Foreigners who were identified with the foreigner icon indicated on the EMR system were excluded.

    1. Data collection

Data were collected by retrospective manual EMR and PACS review by a single investigator from March 2018 to July 2019. AD was initially identified with ICD 10 codes (K57.0-57.9), followed by review by an in- vestigator to exclude cases occurring in sites other than the colon and diverticulitis not diagnosed in the ED. Patient age at the point of diagno- sis of diverticulitis in the ED was obtained by subtracting the date of birth entered in the EMR system from the date of ED diverticulitis diag- nosis. Sex was recorded as presented on the EMR system. Patient past medical history and symptoms and signs were obtained from chart re- view of data entered during the ED visit described by the assigned resi- dent, intern, or consulting specialist. Charlson Comorbidity Index scores were calculated from chart review results [23]. Vital signs of the patient upon ED arrival measured by the triage nurse, in addition to laboratory and CT exam results, were obtained by manually reviewing the EMR system. Personal data, including patient name, address and social secu- rity number, were not obtained. All of the data were collected on a Microsoft Excel spreadsheet (Microsoft, Redmond, WA, USA) on the computer used for EMR review and protected by a password.

    1. Data variables

Patient sex was collected as a dichotomous variable, and male sex was presented because male sex has been previously suggested to be a risk factor for complicated disease [10,20]. Patient age at the point of diagnosis of AD was collected as a continuous variable, as in pre- vious studies [10,19]. Patient past medical history (diabetes, hyper- tension, heart disease, liver disease, renal disease, neurological disability, malignancy, diverticulitis, abdominal surgery), symptoms and signs (duration of pain, anorexia, nausea, vomiting, diarrhea, constipation, hematochezia, fever, abdomen tenderness, rebound ten- derness, guarding) were collected as dichotomous variables. The Charlson comorbidity index [23], a weighted index that accounts for the number and the seriousness of comorbid diseases, was re- corded as a dichotomous variable: 0 versus 1 (greater than 0;

because more than 85% of the total population had a Charlson co- morbidity index less than 2).

Vital signs (blood pressure, respiratory rate, heart rate, body temper- ature) were classified as dichotomous variables. Systolic blood pressure (SBP) less than 100 mmHg was categorized as hypotension as in the quick Sequential Organ Failure Assessment score (qSOFA) [24]. Heart rate, respiratory rate and body temperature were dichotomized based on a heart rate of 90 beats per minute and respiratory rate of 20 breaths per minute according to the systemic inflammatory response syndrome criteria (SIRS) [25]. There were no patients with heart rates less than 50 beats per minute, and only one patient had a respiratory rate less than 16 breaths per minute. Body temperature was categorized based on a temperature of 37.5 ?C, as in previous studies [10,19].

Laboratory results were collected as continuous variables and classi- fied with dichotomous variables based on normal laboratory cutoff values of the hospital with the exception of WBC count and CRP. WBC count and CRP utilized cutoff values of 15,000/ul and 10 mg/dL, respec- tively, based on the results of previous studies [19,26]. There was only one patient with leukocytopenia and hyperkalemia. No patients had hyponatremia according to the normal laboratory range.

Our study collected two CT-related variables: the presence of left colon diverticulitis and the presence of complications. Right-colon di- verticulitis was defined as diverticulitis extending from the cecum to the transverse colon, and left colon diverticulitis was defined as divertic- ulitis extending from the splenic flexure to the sigmoid colon [27]. Com- plicated diverticulitis on CT was defined based on evidence of perforation, abscess, fistula, or colon obstruction [17,18,20,28].

    1. Outcome measures

A clinically important outcome (CIO) of AD was defined if any of the following were present: any cause of death during the hospital stay after admission, any ICU admission during the hospital stay, 7 or more hospi- tal days before discharge, and surgery or invasive intervention (i.e., percutaneous drainage) due to diverticulitis during the hospital stay [17,22,29,30]. Patients included in our study who did not meet the CIO definition were categorized into the nonclinically important outcome (NCIO) group. The primary outcome of our study was a CIO from AD.

    1. Statistical analysis

Categorical variables are presented as numbers and percentages. Continuous variables are reported as medians and interquartile ranges (IQRs). The differences between categorical variables were compared with Pearson’s chi-square test and Fisher’s exact test. Continuous vari- ables were compared with Student’s t-test for normally distributed var- iables and with the Mann-Whitney U test for nonnormally distributed variables.

To assess potential predictors, we conducted univariate analysis. Subsequently, variables with a P-value <=0.20 in univariate analysis were included in Multivariable logistic regression models to identify risk factors predicting CIOs. The cutoff P-value of 0.20 has been fre- quently used for candidate variable selection for prediction models in previous studies [31,32]. The initial model included patient history var- iables that could be obtained by History taking. The results of physical examination, laboratory exam and CT were added to the following models. Model discrimination and calibration of the final model were assessed using the area under the curve (AUC) and the Hosmer- Lemeshow test [33].

Validation of the final model was conducted using 500 randomly drawn bootstrap samples with replacements from the original data [34]. Bootstrapping is a tool used for model validation and is known to be more efficient than data-splitting and cross-validation techniques, and it preserves the sample size, leading to more precision and power [35]. The expected optimism was calculated as the average difference

between the AUC of models when applied to the bootstrap samples and the AUC of the same models when applied to the original data. In addition to computing the optimism-corrected AUC, we aggregated the data generated by bootstrapping to test our prediction model’s dis- crimination and calibration performance [34,36].

The results are presented herein as odds ratios (ORs) with 95% con- fidence intervals (CIs). Statistical analyses were conducted using SAS software, version 9.4 (SAS Institute Inc., Cary, NC, USA). All of the tests were two sided, and P-values <0.05 were considered statistically signif- icant. The study was approved by the Institutional Review Board of the study site (IRB No. 20180918/30-2018-63/103). Informed consent was waived due to the retrospective nature of the study.

  1. Results
    1. Patient demographics

During the study period, 379 patients were diagnosed with AD in the ED. After excluding 36 patients with missing laboratory results and 6 patients with diverticulitis of the duodenum, a total of 337 patients were analyzed, with 63 (18.7%) patients included in the CIO group and 274 (81.3%) patients in the NCIO group (Fig. 1). One patient died during the hospital stay, 5 patients were admitted to the ICU, 16 patients received invasive procedures, and 62 patients had 7 or more hospital days before discharge.

The baseline characteristics of the whole study population are sum- marized in Table 1. Patients with CIOs were significantly older, more commonly had hypertension and renal disease and tended to have had a longer duration of abdominal pain until ED arrival. Symptoms of anorexia, vomiting, and fever were observed more frequently in CIO pa- tients than in NCIO patients, and rebound tenderness was found on physical examination. In addition, the proportion of patients with hypo- tension was greater in patients with CIOs.

The laboratory and CT results of patients with both CIOs and NCIOs are presented in Table 2. WBC and CRP results were higher in the CIO group. Left colon diverticulitis and complications were found more fre- quently on CT results in patients with CIO. Among the total population that was analyzed, 302 (89.6%) patients were diagnosed right-colon di- verticulitis and 35 (10.4%) patients were diagnosed left-colon diverticu- litis. 51.4% (N = 18) of the left-colon diverticulitis population had complications found on CT compared to 10.3% (N = 31) of the right- colon diverticulitis population.

49 (14.5%) patients of the total 337 patients analyzed had complica- tions found on CT and among the 49 patients with complications on CT, 17 (34.7%) patients had NCIOs. Among the 288 patients without compli- cations on CT, 31 (10.8%) patients had CIOs.

    1. Univariate analysis

The results of the univariate analysis for the evaluation of candidate predictors from baseline characteristics, including patient demo- graphics, past medical history, signs and symptoms, and vital signs, are provided in Table 3. Patient age, hypertension, heart disease, renal disease, Charlson comorbidity index equal to or greater than 1, duration of abdominal pain, anorexia, vomiting, fever, rebound tenderness and systolic blood pressure were statistically significantly associated with CIOs in univariate analysis.

The results of the univariate analysis for the evaluation of candidate predictors from laboratory and CT results are provided in Table 4. WBC count, hemoglobin, sodium, blood urea nitrogen, creatinine, glucose, total bilirubin, albumin, alkaline phosphatase, CRP, left-colon diverticu- litis, and complications found on CT were significantly associated with CIOs. Variables with P-values less than 0.20 were used in the multivari- able analysis.

Image of Fig. 1

Fig. 1. Patient enrollment.

    1. Multivariable analysis

Table 5 shows the development of the multivariable logistic regres- sion model. Variables that could be obtained by patient history taking (age, past medical history, duration of abdomen pain, anorexia, vomiting, constipation, and fever) were included in the initial model (Model 1). Age (OR = 1.05, 95% CI = 1.03-1.07), duration of abdominal pain of more than 3 days (OR = 3.59, 95% CI = 1.64-7.84), anorexia (OR

= 3.23, 95% CI = 1.30-8.02) and fever (OR = 2.24, 95% CI = 1.15-4.39)

were predictors of CIOs in the initial model. We subsequently added the results of physical examination (rebound tenderness, guarding, systolic blood pressure, respiratory rate, heart rate, and body temperature) to Model 2. Age (OR = 1.06, 95% CI = 1.04-1.08), duration of abdominal pain of more than 3 days (OR = 3.53, 95% CI = 1.60-7.81), anorexia (OR = 2.76, 95% CI = 1.07-7.11), fever (OR = 2.12, 95% CI =

1.07-4.21), and rebound tenderness (OR = 3.06, 95% CI = 1.56-5.98) were significant predictors of CIOs in Model 2. When laboratory results were added to Model 3, age (OR = 1.05, 95% CI = 1.03-1.07), duration of abdominal pain more than 3 days (OR = 4.36, 95% CI = 1.91-9.96), anorexia (OR = 3.53, 95% CI = 1.26-9.89), rebound tenderness (OR = 3.26, 95% CI = 1.61-6.62), WBC count higher than 15,000/ul (OR = 2.65, 95% CI = 1.13-6.21), and CRP higher than 10 mg/dL (OR = 4.00, 95% CI = 1.97-8.12) were significant predictors of CIOs. In the final model (Model 4), we included patient history, physical examination, labo- ratory exam results, and CT findings in our multivariable logistic regression model. The results showed that increasing age (OR = 1.05, 95% confidence interval CI = 1.02-1.07), duration of abdominal pain of more than 3 days (OR = 3.33, 95% CI = 1.34-8.28), anorexia (OR = 4.31, 95% CI = 1.36-13.65), rebound tenderness (OR = 2.47, 95% CI = 1.15-5.33), WBC

count greater than 15,000/ul (OR = 2.72, 95% CI = 1.09-6.78), C-reactive protein CRP greater than 10 mg/dL (OR = 3.27, 95% CI = 1.51-7.11), and complications found on CT (OR = 8.40, 95% CI = 3.69-19.15) were significant predictors of CIOs.

Discrimination and calibration of the developed models and receiver operating characteristic curves and the Hosmer-Lemeshow goodness-of-fit statistic (?2) are also shown in Table 5. The final model (Model 4) had an AUC of 0.875 (CI = 0.826-0.923). The Hosmer-Lemeshow goodness-of-fit statistic across deciles of risk for

CIOs in the final model was not statistically significant (?2 = 2.969, P = 0.936). When CT results alone were included in the logistic regres- sion model the model (OR = 15.61, 95% CI = 7.78-31.31) had an AUC of 0.723 (CI = 0.659-0.787).

    1. Model validation

Five hundred bootstrap samples with an equal sample size to the original data (N = 337) were generated, resulting in aggregated boot- strap validation data with 168,500 observations. The discrimination, calibration and ROC curves of the developed models when applied to the aggregated bootstrap data (N = 168,500) are shown in Fig. 2. When the aggregated bootstrap data were used for validation of the de- veloped models, the final model (Model 4) had an AUC of 0.874 (CI = 0.872-0.876). The calibration plot of the final model did not show ap- parent deviation from the reference line (Fig. 2).

  1. Discussion

To help provide insight for identifying AD patients with CIOs in the ED, we performed univariate and multivariable logistic regression and developed a prediction model for CIOs with good discrimination and calibration. The prediction model was validated using randomly drawn bootstrap samples. We report that increasing age, duration of ab- dominal pain of more than 3 days, anorexia, rebound tenderness, WBC count greater than 15,000/ul, CRP level greater than 10 mg/dL, and com- plications found on CT are predictors of CIOs in patients diagnosed with diverticulitis in the ED, with right-colon diverticulitis dominant. The prediction model showed higher performance than when CT findings alone were used for prediction of CIOs.

Diverticulitis is a common disease and has been extensively researched in North America and Europe, where left-colon diverticulitis is dominant. However, research considering diverticulitis in the Asian population, in which right-colon diverticulitis is dominant, has been limited. Although the underlying cause of the difference in prevalence of right-colon diverticulitis between Asians and Caucasians is yet to be proven, environmental and genetic causes have been suggested [37-39]. Previous studies evaluating risk factors for poor outcomes of

Table 1

Baseline characteristics of study population

Variables

Total

CIO

NCIO

P value

N %

N %

N %

Total

337 100

63 100

274 100

Sex

Male

213 63.2

36 57.1

177 64.6

0.269

Age, years

Median (IQR)

41 (33-54)

58 (41-73)

39 (32-49)

<0.001

Past medical history

Diabetes

22 6.5

7 11.1

15 5.5

0.151

Hypertension

58 17.2

23 36.5

35 12.8

<0.001

Heart disease

12 3.6

5 7.9

7 2.6

0.053

Liver disease

9 2.7

3 4.8

6 2.2

0.377

Renal disease

9 2.7

6 9.5

3 1.1

0.002

Neurological disability

8 2.4

1 1.6

7 2.6

0.091

Malignancy

3 0.9

2 3.2

1 0.4

0.091

Diverticulitis

30 8.9

7 11.1

23 8.4

0.495

Abdomen surgery

45 13.4

10 15.9

35 12.8

0.514

Charlson comorbidity index >=1

45 13.4

17 27.0

28 10.2

<0.001

Symptoms and Signs

Duration of abdomen pain>3 days

44 13.1

18

28.6

26

9.5

<0.001

Anorexia

31 9.2

14

22.2

17

6.2

<0.001

Nausea

41 12.2

8

12.7

33

12.0

0.886

Vomiting

17 5.0

7

11.1

10

3.6

0.024

Diarrhea

62 18.4

12

19.0

50

18.2

0.883

Constipation

9 2.7

4

6.3

5

1.8

0.067

Hematochezia

5 1.5

1

1.6

4

1.5

1.000

Fever

78 23.1

23

36.5

55

20.1

0.005

Abdomen tenderness

323 95.8

62

98.4

261

95.3

0.482

Rebound tenderness

108 32.0

29

46.0

79

28.8

0.008

Guarding

8 2.4

3

4.8

5

1.8

0.167

Vital signs

SBP < 100 mmHg

10 3.0

5

7.9

5

1.8

0.023

RR > 20 breaths per minute

9 2.7

4

6.3

5

1.8

0.067

HR > 90 beats per min

141 41.8

29

46.0

112

40.9

0.454

BT > 37.5 ?C

36 10.7

11

17.5

25

9.1

0.053

CIO, clinically important outcome. NCIO, nonclinically important outcome. IQR, interquartile range.

SBP, systolic blood pressure. RR, respiratory rate.

HR, heart rate.

BT, body temperature.

patients with diverticulitis in right-colon diverticulitis-dominant popu- lations have suggested that older age, male sex, left-colon diverticulitis, perforation on CT scan, and severe pain are risk factors for severe dis- ease [22,40,41]. However, these studies have been limited by small numbers of admitted patients or outpatients and did not reflect the ED population. Our study is the only one evaluating patients diagnosed with diverticulitis in the ED, where right-colon diverticulitis is domi- nant. Among the 337 patients analyzed, 302 (89.6%) patients had right colon diverticulitis. When compared to patients with right-colon diver- ticulitis, patients with left-colon diverticulitis had higher proportion of patients with complications found on CT and higher proportion of pa- tients with CIOs (10.3% vs. 51.4%, 14.6% vs. 54.3%, respectively). Our study findings, therefore, could be important for Asian populations.

We sought to exclude patients who were diagnosed with diverticu- litis but did not have abdominal CT performed in the ED because the gold standard for diverticulitis is based on CT evaluation, and CT was needed to determine whether complications were present [12]. All pa- tients diagnosed with diverticulitis in the ED during the study period had abdominal CT results. Laboratory tests, including complete blood cell count, electrolyte panel, admission panel, coagulation panel, and CRP, are routinely ordered in patients before CT evaluation in the ED. Thirty-six patients were excluded from the study because they had missing laboratory results, which was considered a general care proto- col violation of the ED evaluation process. This study could be applicable

in study settings in which CT scans and laboratory tests are easily avail- able for evaluating patients suspicious for AD.

Previous studies examining risk factors for the severity of diverticu- litis have traditionally categorized outcomes with CT findings using the Hinchey or Ambrosetti classification [20,42,43]. However, there are lim- itations to this approach because CT is a diagnostic tool and not a clinical outcome. As reported in our study, patients without complications on CT could have poor clinical outcomes, and patients with complications on CT could have favorable clinical outcomes. Our data showed that 31 of 288 patients did not have complications on CT but had CIOs, and 17 of 49 had complications on CT but had NCIOs. To overcome this lim- itation, our study utilized variables such as death, ICU admission, need for surgery or invasive intervention and length of hospital stay to mea- sure CIOs. This approach was used in the study by Juang et al., in which diverticulitis with a severe clinical course was defined by three end- points: need for procedural intervention, admission for more than 7 days and 30-day readmission [30].

Our results showed that age is a significant predictor for CIO. old age

has been proposed by previous studies to be a risk factor for severe dis- ease [19,20,22,26,40] and is in line with previous Asian population- based studies [22,40,41]. As proposed in our study, rebound tenderness on physical examination has been proposed to be a predictor of compli- cated diverticulitis in a systematic review and meta-analysis by Bolkenstein et al. [26]. High WBC counts and CRP levels have also

Table 2

Laboratory and computed tomography results of the study population

Variables

Total

CIO

NCIO

P value

Risk factors

Odds Ratio (95% CI)

P-value

N

%

N

%

N

%

Sex

Total

337

100

63

100

274

100

Laboratory exam

Table 3

Univariate analysis of potential risk factors on clinically important outcome

Male

0.73 (0.42-1.28)

0.270

Age, years

Median (IQR)

1.06 (1.04-1.08)

<0.001

Past medical history

Diabetes

2.16 (0.84-5.54)

0.110

Hypertension

3.93 (2.11-7.33)

<0.001

Heart disease

3.29 (1.01-10.73)

0.049

Liver disease

2.23 (0.54-9.18)

0.265

Renal disease

9.51 (2.31-39.14)

0.002

Neurological disability

0.62 (0.07-5.09)

0.652

Malignancy

8.95 (0.80-100.30)

0.076

Diverticulitis

1.36 (0.56-3.34)

0.496

Abdomen surgery

1.29 (0.60-2.76)

0.515

Charlson comorbidity index >=1 3.25 (1.65-6.41) 0.001

White blood cell count >15,000/uL

50

14.8

18

28.6

32

11.7

0.001

Hemoglobin <12 g/dL

28

8.3

13

20.6

15

5.5

<0.001

Sodium <135 mmol/L

25

7.4

10

15.9

15

5.5

0.013

Potassium <3.5 mmol/L

13

3.9

4

6.3

9

3.3

0.275

Blood urea nitrogen >19 mg/d

30

8.9

16

25.4

14

5.1

<0.001

Creatinine >1.4 mg/dL

12

3.6

7

11.1

5

1.8

0.002

Glucose >110 mg/dL

160

47.5

40

63.5

120

43.8

0.005

Total bilirubin >1.2 mg/dL

107

31.8

27

42.9

80

29.2

0.036

Aspartate aminotransferase

34

10.1

9

14.3

25

9.1

0.220

>40 IU/L

Alanine aminotransferase >40 IU/L

47

13.9

8

12.7

39

14.2

0.751

Albumin <3.3 g/dL

8

2.4

6

9.5

2

0.7

0.001

Alkaline phosphatase >115 IU/L

19

5.6

7

11.1

12

4.4

0.062

International normalized ratio

19

5.6

6

9.5

13

4.7

0.139

> 1.2

C-reactive protein >10 mg/dL

73

21.7

31

49.2

42

15.3

<0.001

Computed tomography Right-colon diverticulitis

302

89.6

44

69.8

258

94.2

<0.001

Left-colon diverticulitis

35

10.4

19

30.2

16

5.8

<0.001

Complication on CT

49

14.5

32

50.8

17

6.2

<0.001

Symptoms and signs

CIO, clinically important outcome. NCIO, nonclinically important outcome. CT, computed tomography.

been frequently reported to be risk factors for severe diverticulitis [20,26,42].

Fortunately, all patients in the study population had records of symptoms and signs assessed in the study due to the utilization of sys- tems and physical examination template. Although the previous litera- ture considering risk factors for AD has not reported a significant association with anorexia and poor outcomes, anorexia was observed to be a significant predictor of CIOs in our multivariable logistic regres- sion model. Acute infections have been suggested to trigger the host’s acute phase response, such as anorexia and lethargy [44]. The discrep- ancy between our results and previous studies might have occurred be- cause most previous studies did not collect data on specific symptoms. Further studies evaluating the underlying physiologic mechanism and association between anorexia and CIO in diverticulitis are needed. Our study found that abdominal pain for more than 3 days was associated with CIOs, in contrast to previous studies by Jaung et al. and Bolkenstein et al., which did not show a significant association between severe dis- ease and the duration of symptoms [30]. Both studies measured the du- ration of symptoms as a continuous variable, while our study measured duration as a categorical variable considering the data distribution.

Previous studies have suggested that patients with right colon diver- ticulitis have a less severe course than patients with left colon divertic- ulitis [8]. Although left colon diverticulitis showed a statistically significant association with CIOs in our univariate analysis, left colon di- verticulitis did not have a statistically significant effect when adjusted in the multivariable logistic regression model. The effect of left colon di- verticulitis could have been mitigated by the effect of CT results on the model because left colon diverticulitis is associated with a larger pro- portion of complications on CT, as mentioned in prior studies [22,40]. As reported in our study, complications such as abscess formation and extraluminal air are known to be predictors of severe disease [17,18,20,28]. Among the whole population, complications on CT were observed in 49 patients, 32 of whom were in the CIO group.

Our study used random samples generated by bootstrapping for validation. Bootstrapping has been frequently used in emergency medicine research to validate prediction models [45,46]. Our predic- tion model showed good discrimination performance in the aggre- gated bootstrap data. The significant Hosmer-Lemeshow goodness- of-fit statistic, calculated when the prediction model was used with

Duration of abdomen pain >3 days

3.82 (1.93-7.53)

0.001

Anorexia

4.32 (2.00-9.33)

<0.005

Nausea

1.06 (0.47-2.43)

0.886

Vomiting

3.30 (1.20-9.04)

0.020

Diarrhea

1.05 (0.52-2.12)

0.883

Constipation

3.65 (0.95-14.00)

0.059

Hematochezia

1.09 (0.12-9.91)

0.940

Fever

2.29 (1.27-4.14)

0.006

Abdomen tenderness

3.09 (0.40-24.02)

0.282

Rebound tenderness

2.11 (1.20-3.69)

0.009

Guarding

2.69 (0.63-11.57)

0.184

Vital signs

Systolic blood pressure < 100 mmHg 4.64 (1.30-16.54) 0.018

Respiratory rate > 20 breaths per min 3.65 (0.95-14.00) 0.059

Heart rate > 90 beats per min 1.23 (0.71-2.14) 0.455

Body Temperature > 37.5 ?C 2.11 (0.98-4.55) 0.058

IQR, interquartile range.

95% CI, 95% confidence interval.

the bootstrap data, is suspected to be due to the large sample size of the aggregated bootstrap data.

    1. Limitations

Our study was limited to an ED of a single center. Patients discharged from the ED or ward could not be followed up due to the retrospective

Table 4

Univariate analysis of laboratory and computed tomography findings regarding clinically important outcomes

Risk factors Odds Ratio (95% CI) P-value Laboratory exam

White blood cell count >15,000/uL 3.03 (1.57-5.85) 0.001

Hemoglobin <12 g/dL 4.49 (2.01-10.01) <0.005

Sodium <135 mmol/L 3.26 (1.39-7.65) 0.007

Potassium<3.5 mmol/L 2.00 (0.60-6.70) 0.263

Blood urea nitrogen >19 mg/d 6.32 (2.89-13.81) <0.001

Creatinine >1.4 mg/dL 6.73 (2.06-21.96) 0.002

Glucose >110 mg/dL 2.23 (1.27-3.93) 0.005

Total bilirubin >1.2 mg/dL 1.82 (1.04-3.19) 0.037

Aspartate aminotransferase >40 IU/L 1.66 (0.73-3.76) 0.224

Alanine aminotransferase >40 IU/L 0.88 (0.39-1.98) 0.751

Albumin <3.3 g/dL 14.30 (2.82-72.66) 0.001

Alkaline phosphatase >115 IU/L 2.73 (1.03-7.24) 0.044

International normalized ratio > 1.2 2.11 (0.77-5.80) 0.146

C-reactive protein >10 mg/dL 5.35 (2.96-9.69) <0.001 Computed tomography findings

Left-colon diverticulitis 6.96 (3.33-14.56) <0.001

Complication on CT 15.61 (7.78-31.31) <0.001

CT, computed tomography. 95% CI, 95% confidence interval.

Table 5

Predictors of clinically important outcomes by multivariable analysis

Models

? Coefficient

OR (95% CI)

AUC (95% CI)

HL, ?2 (p-value)

Model 1: patient history

0.779 (0.711-0.846)

7.52 (0.48)

Age

0.05

1.05 (1.03-1.07)

Duration of abdomen pain >3 days

1.28

3.59 (1.64-7.84)

Anorexia

1.17

3.23 (1.30-8.02)

Fever

0.81

2.24 (1.15-4.39)

Model 2: model 1 factors plus physical examination

0.805 (0.741-0.869)

5.34 (0.72)

Age

0.06

1.06 (1.04-1.08)

Duration of abdomen pain >3 days

1.26

3.53 (1.60-7.81)

Anorexia

1.02

2.76 (1.07-7.11)

Fever

0.75

2.12 (1.07-4.21)

Rebound tenderness

1.12

3.06 (1.56-5.98)

Model 3: model 2 factors plus laboratory exams

0.841 (0.789-0.893)

18.69(0.12)

Age

0.05

1.05 (1.03-1.07)

Duration of abdomen pain >3 days

1.47

4.36 (1.91-9.96)

Anorexia

1.26

3.53 (1.26-9.89)

Rebound tenderness

1.18

3.26 (1.61-6.62)

White blood cell count >15,000/uL

0.97

2.65 (1.13-6.21)

C-reactive protein>10 mg/dL

1.39

4.00 (1.97-8.12)

Model 4: model 3 factors plus CT findings

0.875 (0.826-0.923)

2.97 (0.94)

Age

0.01

1.05 (1.02-1.07)

Duration of abdomen pain >3 days

0.46

3.33 (1.34-8.28)

Anorexia

0.59

4.31 (1.36-13.65)

Rebound tenderness

0.39

2.47 (1.15-5.33)

White blood cell count >15,000/uL

0.47

2.72 (1.09-6.78)

C-reactive protein >10 mg/dL

0.40

3.27 (1.51-7.11)

Complication on CT

0.42

8.40 (3.69-19.15)

CT, computed tomography.

95% CI, 95% confidence interval.

AUC, area under the receiver operator characteristic curve.

HL, Hosmer-Lemeshow test.

nature of the study. Discharged patients could have had poor clinical outcomes and could have been admitted to hospitals other than this study site. Selection bias also could have occurred because our popula- tion only included patients who had CT and laboratory results. Patients with less severe disease or patients with repetitive episodes of divertic- ulitis could have been discharged or admitted without CT or laboratory exams.

Measurement bias cannot be excluded because EMR chart review for the extraction of data and categorization to study outcomes were per- formed by a single investigator. In addition, due to the retrospective na- ture of the study, there were no predefined clinical protocols, and information regarding past medical history, symptoms and signs was dependent on medical records described by the duty ED physician and could have been omitted if not obtained or correctly recorded by the

Image of Fig. 2

Fig. 2. Discrimination and calibration plot of the prediction models for clinically important outcomes in acute diverticulitis in the validation dataset using bootstrap samples

  1. Discrimination performance plot
  2. Calibration performance plot

ED physician during the patient’s ED stay. All patients in the study pop- ulation had records of symptoms and signs assessed in the study due to the utilization of review of systems and physical examination template. The default value for all signs and symptoms in the template are marked as negative and changed by the ED physician to positive if a sign or symptom is found during patient assessment. There is a risk of symp- toms and signs being marked falsely negative if not changed to positive by the ED physician. Pain scores were not included in the ED review of systems and physical examination template. Pain scores of patients were not evaluated because of many missing values. The definition of CIO used in our study has limitations because 62 of 63 patients with CIOs had more than 7 hospital days. Hospital days could have been in- creased due to factors other than diverticulitis, such as hospital days for additional exams unrelated to diverticulitis, family issues, or time spent finding a nursing home facility. One patient in the CIO group had 6 hospital days and was categorized into the CIO group because the patient underwent percutaneous drainage due to diverticulitis.

Our logistic regression model has limitations in that it has not been externally validated. Although we sought to overcome this limitation using aggregated bootstrapped data, this approach has limitations be- cause the bootstrapped data are sampled from the original data. Further studies are needed to validate the prediction model. Due to limitations in our study design, our prediction model for CIOs of AD has limitations in representing the general population.

  1. Conclusion

We developed a prediction model for CIOs in patients diagnosed with AD with good discrimination and calibration that can aid clinical decision making in the ED. The model showed good performance in bootstrap validation data and could be used in emergency department populations where right-colon diverticulitis is dominant.

Funding acknowledgment

None.

Credit author statement

All authors contributed to this study; SKW Lee analyzed the dataset and drafted the manuscript. HJ Lee, GJ Suh, and DJ Park contributed to specific review and add comments. All authors contributed to revision of the manuscript. Shin SD designed the study and has responsible for the study.

Declaration of Competing Interest

Nothing to declare is in this paper.

References

  1. Stollman N, Raskin JB. Diverticular disease of the colon. The Lancet. 2004;363(9409):

631-9. https://doi.org/10.1016/S0140-6736(04)15597-9.

  1. Jacobs DO. Diverticulitis. New England Journal of Medicine. 2007;357(20):2057-66. https://doi.org/10.1056/NEJMcp073228.
  2. Wheat CL, Strate LL. Trends in hospitalization for diverticulitis and diverticular bleeding in the United States from 2000 to 2010. Clinical Gastroenterology and Hepatology. 2016;14(1) (96-103. e1).
  3. Reisman Y, Ziv Y, Kravrovitc D, Negri M, Wolloch Y, Halevy A. Diverticulitis: the ef- fect of age and location on the course of disease. Int J Colorectal Dis. 1999;14 (4-5):250-4.
  4. Hall JF, Roberts PL, Ricciardi R, Read T, Scheirey C, Wald C, et al. Long-term follow-up after an initial episode of diverticulitis: what are the predictors of recurrence? Dis- eases of the colon & rectum. 2011;54(3):283-8.
  5. Lim DR, Kuk JC, Shin EJ, Hur H, Min BS, Lee KY, et al. Clinical outcome for manage- ment of colonic diverticulitis: characteristics and surgical factor based on two insti- tution data at South Korea. Int J Colorectal Dis. 2020;35(9):1711-8.
  6. Law W, Lo C, Chu K. Emergency surgery for colonic diverticulitis: differences be- tween right-sided and left-sided lesions. Int J Colorectal Dis. 2001;16(5):280-4.
  7. Chia JG, Wilde CC, Ngoi SS, Goh PM, Ong CL. Trends of diverticular disease of the large bowel in a newly developed country. Diseases of the colon & rectum. 1991; 34(6):498-501.
  8. Yamada T, Alpers DH, Kalloo AN, Kaplowitz N, Owyang C, Powell DW. Textbook of gastroenterology. John Wiley & Sons; 2011.
  9. Longstreth G, Iyer R, Chu LH, Chen W, Yen L, Hodgkins P, et al. Acute diverticulitis: demographic, clinical and Laboratory features associated with computed tomogra- phy findings in 741 patients. Aliment Pharmacol Ther. 2012;36(9):886-94.
  10. Ambrosetti P, Grossholz M, Becker C, Terrier F, Morel P. Computed tomography in acute left colonic diverticulitis. Br J Surg. 1997;84(4):532-4.
  11. Sartelli M, Weber DG, Kluger Y, Ansaloni L, Coccolini F, Abu-Zidan F, et al. 2020 up- date of the WSES guidelines for the management of acute colonic diverticulitis in the emergency setting. World Journal of Emergency Surgery. 2020;15(1):1-18.
  12. Chabok A, Pahlman L, Hjern F, Haapaniemi S, Smedh K, Group AS. Randomized clin- ical trial of antibiotics in acute uncomplicated diverticulitis. Br J Surg. 2012;99(4): 532-9. https://doi.org/10.1002/bjs.8688.
  13. Shabanzadeh DM, Wille-Jorgensen P. Antibiotics for uncomplicated diverticulitis. Cochrane Database Syst Rev. 2012;11:CD009092.
  14. Hall J, Hardiman K, Lee S, Lightner A, Stocchi L, Paquette IM, et al. The American So- ciety of Colon and Rectal Surgeons clinical practice guidelines for the treatment of left-sided colonic diverticulitis. Diseases of the Colon & Rectum. 2020;63(6):728-47.
  15. Etzioni DA, Chiu VY, Cannom RR, Burchette RJ, Haigh PI, Abbas MA. Outpatient treat- ment of acute diverticulitis: rates and predictors of failure. Diseases of the Colon & Rectum. 2010;53(6):861-5.
  16. Lorimer JW, Doumit G. Comorbidity is a major determinant of severity in acute di- verticulitis. The American Journal of Surgery. 2007;193(6):681-5. https://doi.org/ 10.1016/j.amjsurg.2006.10.019.
  17. Ambrosetti P. Acute diverticulitis of the left Colon: value of the initial CT and timing of elective colectomy. J Gastrointest Surg. 2008;12(8):1318. https://doi.org/10.1007/ s11605-008-0489-8.
  18. Van de Wall B, Draaisma W, Van der Kaaij R, Consten E, Wiezer M, Broeders I. The value of inflammation markers and body temperature in acute diverticulitis. Colo- rectal Dis. 2013;15(5):621-6.
  19. Makela JT, Klintrup K, Takala H, Rautio T. The role of C-reactive protein in prediction of the severity of acute diverticulitis in an emergency unit. Scand J Gastroenterol. 2015;50(5):536-41.
  20. Bollom A, Austrie J, Hirsch W, Nee J, Friedlander D, Ellingson K, et al. Emergency de- partment burden of diverticulitis in the USA, 2006-2013. Dig Dis Sci. 2017;62(10): 2694-703.
  21. Kim SY, Oh TH, Seo JY, Jeon TJ, Seo DD, Shin WC, et al. The clinical factors for predicting severe diverticulitis in Korea: a comparison with Western countries. Gut and liver. 2012;6(1):78.
  22. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prog- nostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373-83. https://doi.org/10.1016/0021-9681(87)90171-8.
  23. Seymour CW, Liu VX, Iwashyna TJ, Brunkhorst FM, Rea TD, Scherag A, et al. Assess- ment of clinical criteria for sepsis: for the third international consensus definitions for Sepsis and septic shock (Sepsis-3). Jama. 2016;315(8):762-74.
  24. Dellinger RP, Levy MM, Rhodes A, Annane D, Gerlach H, Opal SM, et al. Surviving Sepsis campaign: international guidelines for management of severe sepsis and sep- tic shock, 2012. Intensive Care Med. 2013;39(2):165-228.
  25. Bolkenstein H, van de Wall B, Consten E, Broeders I, Draaisma W. Risk factors for complicated diverticulitis: systematic review and meta-analysis. Int J Colorectal Dis. 2017;32(10):1375-83.
  26. Lee K-y, Lee J, Park YY, Kim Y, Oh ST. Difference in clinical features between right- and left-sided acute colonic diverticulitis. Sci Rep. 2020;10(1):1-6.
  27. Feingold D, Steele SR, Lee S, Kaiser A, Boushey R, Buie WD, et al. Practice parameters for the treatment of sigmoid diverticulitis. Diseases of the Colon & Rectum. 2014;57 (3):284-94.
  28. Kechagias A, Rautio T, Kechagias G, Makela J. The role of C-reactive protein in the prediction of the clinical severity of acute diverticulitis. Am Surg. 2014;80(4):391-5.
  29. Jaung R, Kularatna M, Robertson JP, Vather R, Rowbotham D, MacCormick AD, et al. Uncomplicated acute diverticulitis: identifying risk factors for Severe outcomes. World J Surg. 2017;41(9):2258-65.
  30. Scheuermeyer FX, Wong H, Yu E, Boychuk B, Innes G, Grafstein E, et al. Development and validation of a prediction rule for early discharge of low-risk emergency depart- ment patients with potential ischemic chest pain. Canadian Journal of Emergency Medicine. 2014;16(2):106-19.
  31. Antman EM, Cohen M, Bernink PJ, McCabe CH, Horacek T, Papuchis G, et al. The TIMI risk score for unstable angina/non-ST elevation MI: a method for prognostication and therapeutic decision making. Jama. 2000;284(7):835-42.
  32. Lemeshow S, Hosmer Jr DW. A review of goodness of fit statistics for use in the de- velopment of logistic regression models. Am J Epidemiol. 1982;115(1):92-106.
  33. Steyerberg EW, Harrell FE. Prediction models need appropriate internal, internal- external, and external validation. J Clin Epidemiol. 2016;69:245-7.
  34. Efron B, Tibshirani RJ. An introduction to the bootstrap. CRC press; 1994.
  35. Alba AC, Agoritsas T, Walsh M, Hanna S, Iorio A, Devereaux P, et al. Discrimination and calibration of clinical prediction models: users’ guides to the medical literature. Jama. 2017;318(14):1377-84.
  36. Hjern F, Johansson C, Mellgren A, Baxter N, Hjern A. Diverticular disease and migration-the influence of acculturation to a Western lifestyle on diverticular dis- ease. Aliment Pharmacol Ther. 2006;23(6):797-805.
  37. Miura S, Kodaira S, Shatari T, Nishioka M, Hosoda Y, Hisa K. Recent trends in diver- ticulosis of the right colon in Japan. Diseases of the colon & rectum. 2000;43(10): 1383-9.
  38. Lee Y-S. Diverticular disease of the large bowel in Singapore. Diseases of the colon & rectum. 1986;29(5):330-5.
  39. Park N-S, Jeen YT, Choi HS, Kim ES, Kim YJ, Keum B, et al. Risk factors for severe di- verticulitis in computed tomography-confirmed acute diverticulitis in Korea. Gut and liver. 2013;7(4):443.
  40. Yoo T, Yang KH, Kim J, Park I, Cho H, Gwak G, et al. Predictive factors affecting the clinical course of patients with diverticulitis: who needs hospital management? An- nals of coloproctology. 2018;34(1):23-8.
  41. Tursi A, Brandimarte G, Giorgetti G, Elisei W, Maiorano M, Aiello F. The clinical pic- ture of uncomplicated versus complicated diverticulitis of the colon. Dig Dis Sci. 2008;53(9):2474-9. https://link.springer.com/content/pdf/10.1007%2Fs10620-007- 0161-2.pdf.
  42. Hall JF, Roberts PL, Ricciardi R, Marcello PW, Scheirey C, Wald C, et al. Colonic diver- ticulitis: does age predict severity of disease on CT imaging? Diseases of the Colon & Rectum. 2010;53(2):121-5.
  43. Langhans W. Signals generating anorexia during acute illness: symposium on ‘eat- ing, illness and the gut: is there disorder in the house?’. Proceedings of the Nutrition Society. 2007;66(3):321-30.
  44. Barrett TW, Martin AR, Storrow AB, Jenkins CA, Harrell Jr FE, Russ S, et al. A clinical prediction model to estimate risk for 30-day adverse events in emergency depart- ment patients with symptomatic atrial fibrillation. Ann Emerg Med. 2011;57(1): 1-12.
  45. Hess EP, Brison RJ, Perry JJ, Calder LA, Thiruganasambandamoorthy V, Agarwal D, et al. Development of a clinical prediction rule for 30-day cardiac events in emer- gency department patients with chest pain and possible acute coronary syndrome. Annals of emergency medicine. 2012;59(2) (115-25. e1).