Using Patient Characteristics to Predict Usefulness of Abdominal Computed Tomography in Children
Objective: The study goal was to determine whether clinical symptoms, physical findings, or laboratory values predict the usefulness of abdominal computed tomography (CT) scans in children.
Methods: We conducted a retrospective review of pediatric patients who received an abdominal CT scan between June 2009 and November 2011 at an urban medical center. A panel of pediatric hospitalists independently categorized each CT read as: (1) necessary for diagnosis; (2) unnecessary but helpful for diagnosis; or (3) neither necessary nor helpful for diagnosis. Two multiple logistic regression models examined 21 clinical variables to assess their ability to differentiate between: (1) necessary and unnecessary CT scans; and (2) helpful and unhelpful CT scans.
Results: A total of 399 CT scans were analyzed. Seventy (18%) of these were categorized as necessary, 103 (26%) as unnecessary but helpful, and 226 (57%) as neither necessary nor helpful. Three variables were strongly correlated with necessary CT scans: leukocytosis, peritoneal signs, and male gender. The probability of a CT scan being necessary was 57% in patients with all 3 findings and 8% in those with none. Three variables were also strongly correlated with unnecessary but helpful CT scans: history of abdominal surgery, tachypnea, and leukocytosis. The probability of a CT scan being helpful was 84% in patients with all 3 findings and 35% in those with none of the 3 findings.
Conclusions: The majority of abdominal CT scans were unnecessary and unhelpful. Knowing which clinical variables correlate strongly with necessary or helpful abdominal CT scans may assist clinicians in deciding whether to order this test; however, the predictive power of these variables remains relatively low.
Radiation exposure from computed tomography (CT) scans has been a popular topic in both the adult and pediatric literature in recent years.1,2 Since 1980, the use of CT scans has increased >20-fold in the United States; in 2007, >4 million pediatric CT scans were ordered.3,4 As CT scans have become a common diagnostic imaging modality, concerns regarding long-term consequences of radiation exposure have arisen. The attributable lifetime cancer mortality risk from 1 abdominal CT scan is estimated to be 0.14% at birth and 0.065% at 18 years of age.1,5,6 Furthermore, these data do not include nonfatal CT scan–related cancer incidence, which may be as high as 1 in 900 of all patients undergoing CT scans.7 Although population risks are debated, it has been estimated that up to 2% of current US cancers are related to CT radiation.1
Pediatric patients are at increased risk for developing fatal cancer as a consequence of radiation exposure compared with adults.1–3,5,6,8,9 However, although risks associated with radiation exposure among pediatric patients are generally accepted, there is little clinician consensus regarding clinical indications for the use of CT scans in pediatric patients, especially for abdominal pain. Although the use of CT scans to evaluate abdominal pain in adults more than doubled from 1995 to 2005,10 studies focusing on the accuracy of abdominal CT scans performed in children with abdominal pain are scarce and have generally focused on appendicitis.7,9,10
Plain radiographs have limited value in abdominal pain,9,11 but 2 prospective adult studies have demonstrated that CT scan findings can have a significant effect on diagnosis and may change management in 46% to 60% of patients with abdominal pain.12,13 Similar pediatric data are not available, and extrapolating results of adult studies to the pediatric population is difficult because the differential diagnosis of abdominal pain in adults differs from that in children. Diseases for which CT scans are helpful in diagnosis, such as diverticulitis and oncologic processes, are more prevalent in the adult population, whereas conditions not requiring a CT scan, such as gastroenteritis, constipation, and functional abdominal pain, are more commonly encountered in pediatrics.14,15 Therefore, it is likely that the pretest probability of a CT scan changing clinical management is higher for an adult with abdominal pain than it is for a child.
The goal of the current study was to assess whether clinical information can predict CT scan results that will affect clinical decision-making in children.
This study was a retrospective chart review of pediatric patients who received an abdominal CT scan for abdominal pain at Golisano Children’s Hospital (GCH) at the University of Rochester Medical Center between June 2009 and November 2011. The study was approved by the University of Rochester Medical Center institutional review board.
GCH is a tertiary care children’s hospital and regional referral center located in Rochester, New York. General inpatient pediatric care is largely provided by pediatric residents, pediatric hospitalists, and pediatric subspecialists. Pediatric emergency department care is provided by a mix of board-certified and nonboard-certified pediatric emergency physicians as well as pediatric and emergency medicine residents.
Inclusion criteria for this study were age <19 years, documented International Classification of Diseases, Ninth Revision, procedure code 88.01 for abdominal CT in either the emergency department or inpatient units, and either a chief complaint of abdominal pain documented in the patient’s chart or an entry in the medical record indicating that the CT scan was obtained to evaluate abdominal pain. CT scans performed for multiple indications were included as long as 1 of the recorded indications was abdominal pain.
We excluded patients with complicating underlying illness that would place them at risk for findings different from the general pediatric population. These illnesses included abdominal pain in the setting of trauma; history of oncologic, end-stage renal, or chronic gastrointestinal disease; history of pancreatitis; abdominal surgery <3 weeks before the CT scan; and, if the abdominal CT scan was done at <2 years of age, gestational age at birth ≤35 weeks. Abdominal CT scans performed to follow up a previous scan were also excluded.
Eligible patients were identified by the Office of Clinical Practice Evaluation at the University of Rochester Medical Center by searching the hospital administrative data set for patients who had abdominal CT scans performed. Information regarding patients meeting study criteria was then collected from electronic medical records or paper charts located on-site at GCH. Charts were reviewed in reverse chronological order, starting with November 2011. Statistical testing was completed periodically; once our analysis reached 399 charts (June 2009), we had sufficient power to detect a significant difference in our variables of interest, and we stopped our data collection.
For each patient, the abstractors recorded the chief complaint, the formal CT reading by the attending radiologist (either a pediatric or adult radiologist), and all final discharge diagnoses as they appeared on the Office of Clinical Practice Evaluation data set.
Twenty-one demographic and clinical variables were recorded by the abstractors as either present or absent in the 48 hours before the CT scan. The variables recorded were similar to those used in another study evaluating abdominal imaging modalities.15 Variables included demographic information (age, gender), health information (previous gastrointestinal diagnoses [reflux, peptic ulcer disease, constipation] that would not typically require a CT scan, history of abdominal surgery, ventriculoperitoneal shunt), descriptors of present illness (emesis, diarrhea, gastrointestinal bleed, dysuria, flank pain, hematuria, foreign body ingestion), vital signs (fever, tachycardia, tachypnea), physical examination findings (abdominal tenderness, peritoneal signs, distension, abnormal bowel sounds), and laboratory findings (leukocytosis, bandemia). When there was a discrepancy among providers describing a particular variable, documented values from the most senior physician were used.
Tachypnea, tachycardia, and leukocytosis were defined as a value above the normal, age-specific ranges as published in The Harriet Lane Handbook (18th edition).16 Hematuria was defined as >3 red blood cells per high-power field in the urine identified by microscopy or a urine test strip positive for blood; bandemia as >5% bands present on the differential analysis of the white blood cell count; and fever as a temperature ≥38.0°C. Gastrointestinal bleeding was recorded as present if it was reported by history or found on fecal occult blood testing.
After reviewing expert opinion from a senior pediatric radiologist and a senior pediatric nephrologist regarding conditions in which a CT scan may be necessary or helpful, 2 attending pediatric hospitalists who were blinded to the demographic data and clinical variables independently examined the CT reading and discharge diagnosis(es) for each patient. They placed each CT scan into 1 of 3 categories: (1) necessary and helpful for diagnosis or management; (2) unnecessary but helpful for diagnosis or management; and (3) unnecessary and unhelpful for diagnosis or management. CT scans in which there was disagreement between the 2 hospitalists (12 of 430 studies [2.8%]) were sent to a third senior pediatric hospitalist for final categorization.
Some of the most common CT scans categorized as necessary were those demonstrating evidence for appendicitis, abscess, and inflammatory bowel disease. Based on expert opinion from a pediatric nephrologist, CT scans showing renal stones were categorized as necessary if the stone was concurrent with obstruction. At our institution, a CT scan is performed to diagnose appendicitis if there is a clinical suspicion of appendicitis and an ultrasound is either inconclusive or not available; therefore, CT scans diagnosing appendicitis were considered necessary. The raters generally assumed that a CT scan conducted to “rule out” a diagnosis was unnecessary if the results did not validate that diagnosis. For example, a CT scan conducted to evaluate a patient for appendicitis who was not subsequently diagnosed with appendicitis was considered to have been unnecessary.
A CT scan was classified as helpful if it demonstrated positive findings that helped in diagnosis or management but were not required for diagnosis or management (ie, could have been diagnosed by clinical criteria, laboratory testing, ultrasound). The diagnoses most commonly identified by a helpful CT scan were pyelonephritis, ovarian cyst, and cholecystitis.
All variables were coded as dichotomous variables (present/absent; male/female; age ≤13 years/>13 years). Foreign body ingestion did not occur in any of our subjects and was therefore excluded, leaving 20 variables for analysis. For demographic, vital sign, and laboratory variables, if the presence or absence of the variable could not be identified from the chart review, the variable was considered missing. For all other variables, the default was to consider it absent unless otherwise elicited from chart review. For example, dysuria was considered absent unless otherwise documented.
The data were analyzed by using a multiple logistic regression model. Two separate analyses were performed. The first analysis compared necessary with unnecessary CT scans and the second compared helpful with unhelpful CT scans. In both analyses, the predictors were the 20 clinical variables. PROC LOGISTIC in SAS version 9.2 (SAS Institute, Inc, Cary, NC) was used for both model fitting and variable selection.
A total of 1055 patients were identified as having an abdominal CT scan performed in the GCH pediatric emergency department or on the general inpatient wards between June 1, 2009, and November 30, 2011. Of these, 430 patients met inclusion criteria and were categorized. Thirty-one subjects were excluded due to missing values for 1 or more of the clinical variables. In the remaining 399 subjects, 70 (18%) CT scans were categorized as necessary, 103 (26%) were categorized as helpful, and 226 (57%) were categorized as unhelpful.
Table 1 presents the basic demographic information for our study subjects as well as their most common chief complaints. Table 2 displays the frequency of the clinical variables within each CT category. Several variables had very small nonzero counts.
Necessary Versus Unnecessary CT Scans
Regression analysis identified 3 covariates as increasing the probability of having a necessary CT scan: leukocytosis, peritoneal signs, and male gender. Table 3 displays the sensitivities, specificities, positive likelihood ratios (PLR), negative likelihood ratios (NLR), and odds ratio (OR) estimates of the selected model for these variables. Male gender was the covariate with the highest sensitivity (47%); leukocytosis had the strongest specificity (81%), PLR (2.35), NLR (0.69), and OR (3.28). Figure 1 displays the receiver operating characteristic (ROC) curve created by using combinations of sensitivities and specificities for these 3 covariates. The area under the ROC curve was only 0.69 (95% CI: 0.61–0.76), indicating that the accuracy of the variables in predicting necessary CT scans was fairly low. The probability of a CT scan being necessary to make a diagnosis was estimated to be 57% in patients with all 3 clinical findings present and 8% in those with none of the 3. Conversely, 92% of CT scans performed in female subjects without leukocytosis and peritoneal signs were unnecessary. Table 4 reports the contingency data for gender and age. The likelihood of a male subject aged >13 years undergoing a necessary CT scan was 30% (95% CI: 19–41), whereas the likelihood of a female subject aged >13 years undergoing a necessary CT scan was 14% (95% CI: 9–18). The difference in gender proportion was accounted for in the regression model.
Helpful Versus Unhelpful CT Scans
Three covariates were identified as increasing the probability of a CT scan being helpful: history of abdominal surgery, tachypnea, and leukocytosis. Table 3 displays the sensitivities, specificities, PLR, NLR, and OR estimates of the selected model for these variables. Leukocytosis had the highest sensitivity (31%) and NLR (0.84); history of abdominal surgery had the highest specificity (94%), PLR (2.11), and OR (2.37). All 3 covariates increased the probability of having a helpful CT reading. Figure 2 shows the ROC curve generated for this model. The area under the ROC curve was 0.60 (95% CI: 0.55–0.65), indicating poor accuracy. Overall, the probability of a CT scan being helpful was 84% in patients with all 3 findings and 35% in those with none of the 3 findings.
Several clinical variables were strongly correlated with clinically useful CT results. In the first analysis (necessary versus unnecessary), leukocytosis, peritoneal signs, and male gender were significantly correlated with findings on CT scans that were necessary for diagnosis or management. Our results provide 3 primary conclusions. First, CT scans performed in female patients with abdominal pain but without leukocytosis and peritoneal signs were either normal or identified a disease that usually (92% of the time) did not require a CT scan to diagnose. Specifically, although the distribution of CT categories is similar for male and female subjects ≤13 years of age, in the group aged >13 years, female subjects were much more likely to have an unnecessary CT scan. These data suggest that a period of observation or trial of therapy for more common diseases may be warranted before obtaining a CT scan, particularly in the adolescent female population. Second, there is little evidence to support a strong pretest probability that an abdominal CT scan will be necessary. Of the scans analyzed in this study, we identified only 18% as being necessary. Third, the use of patient characteristics to predict necessary CT scans was generally not very accurate. Although we found that a high number of abdominal CT scans for abdominal pain may be unnecessary, the decision to obtain an abdominal CT scan should be based more on overall risk than on any specific combination of patient characteristics. However, the presence of leukocytosis and peritoneal signs did seem to increase the likelihood that an abdominal CT scan may be needed to make a definitive diagnosis, especially in boys. In addition, when a patient presents with abdominal pain for which a CT scan is requested, the likelihood of that CT scan being necessary for diagnosis or clinical decision-making in female subjects was one-half that of male subjects. This finding is likely attributable in part to unique anatomic differences and a broader differential diagnosis in female subjects which includes many diagnoses that do not require a CT scan. Although it can be debated whether a CT scan is necessary for diagnosis or management in different clinical circumstances, the appropriateness of our categorization is supported by the high level of agreement among the independent raters (κ = 0.096).
The second analysis (helpful versus unhelpful) revealed that leukocytosis was again significant, along with a history of abdominal surgery and tachypnea. One difficulty of placing CT scans into this category was determining whether certain findings were incidental or pathologic. The raters attempted to alleviate this situation somewhat by assuming that a finding was a pathologic cause of the patient’s abdominal pain if it was included as 1 of the patient’s final diagnoses. For example, a CT scan indicating a hemorrhagic adnexal cyst was considered helpful if the cyst was listed as a final diagnosis but incidental if it was not. However, there is a possibility that some cases were missed secondary to lack of inclusion in the discharge diagnoses.
Similar to the first analysis, combining the 3 significant covariates of leukocytosis, history of abdominal surgery, and tachypnea did not strongly influence the pretest probability of obtaining a helpful CT scan. In addition, the majority of variables analyzed were not significant in predicting CT utility. This finding may be partially explained by the low incidence of several of these variables (eg, gastrointestinal bleed, hematuria, abdominal distension).
Although our data indicate that the clinical patient information selected in this study was not highly predictive of the diagnostic utility of an abdominal CT scan, more detailed studies examining the variables that were strongly correlated with necessary or helpful CT scans could define further their clinical usefulness, either independently or in combination.
Although neither model indicated age alone to be correlated with necessary or helpful CT scans, further stratification according to age, specifically within the cohort aged <13 years, might show different results. A recent study examining negative appendectomy rates in children found that utility of advanced imaging modalities such as CT scans can vary by age.17 A future study should use a larger sample of patients, which will allow examination of several of our more infrequent variables such as gastrointestinal bleeding and abdominal distension. This future study should also have the raters evaluate the entire chart to better categorize the CT scan.
Our study had several limitations. First, the decision to classify CT scans performed to “rule out” serious pathology as unnecessary can be debated. Although a normal study may provide benefit in terms of triage, reassurance, and diagnosis, the goal of our study was to determine whether patient characteristics could predict pathology that necessitates an abdominal CT scan. If a normal study was identified, the patient’s clinical findings were predictive of an unnecessary, rather than necessary, CT scan. For example, a physician may order a CT scan to exclude a devastating illness and confirm a more benign diagnosis. The CT scan may have been necessary to provide reassurance and/or to avoid further diagnostic evaluation; however, we categorized such a CT scan as unnecessary because we wanted to identify whether, in such a situation, the patient characteristics alone could have predicted the normal CT result, avoiding the need for the “rule out” test. Prospective observational studies would be required to better assess the factors associated with the optimal use of CT scans, but we were unable to support more comprehensive review.
Second, certain variables are subjective and have the potential to be overdocumented or underdocumented by physicians. Third, CT reads were finalized by only 1 attending radiologist without corroboration by a second. This method, however, is standard practice and replicates the “real world” with which a physician would be confronted. Fourth, in our data extraction, we did not record whether an ultrasound had been performed on the patient. It is possible that, if a CT scan was ordered because an ultrasound was unavailable, our data may have been skewed because some unnecessary CT scans may have been avoided. Fifth, our method of CT categorization could be criticized as excluding pertinent negatives from being read as “helpful” or “necessary.” Sixth, our method of determining when CT findings that would otherwise be described as incidental were pathologic was not well defined and does introduce the possibility of bias; however, the high level of agreement between the first 2 raters provides some reassurance. Finally, our study did not have sufficient power to detect the significance of infrequent variables.
Our study demonstrated that the majority of these abdominal CT scans were unhelpful, exposing children to excessive CT-related radiation and that certain clinical variables may help to predict the utility of CT scans in these patients. A prospective study would be helpful to identify the risk-benefit ratio of an abdominal CT scan in the pediatric population.
The authors thank Dr Mary Fraga for her help with data collection.
The authors had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the analysis.
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: This project was supported by the University of Rochester CTSA award UL1 RR024160 from the National Center for Research Resources and the National Center for Advancing Translational Sciences of the National Institutes of Health.
- confidence interval
- computed tomography
- Golisano Children’s Hospital
- negative likelihood ratios
- odds ratio
- positive likelihood ratios
- receiver operating characteristic
- Brody AS,
- Frush DP,
- Huda W,
- Brent RL
- Hall EJ,
- Brenner DJ
- 4.↵IMV 2006 CT Market Summary Report. Des Plains, IL: IMV Medical Information Division; 2006.
- Broder JS,
- Hollingsworth CL,
- Miller CM,
- Meyer JL,
- Paulson EK
- Frush DP,
- Donnelly LF,
- Rosen NS
- Manson D
- 10.↵ICRP 2007 Recommendations of the International Commission on Radiological Protection. ICRP Publication 103; Ann ICRP 37 (2–4)
- Scholer SJ,
- Pituch K,
- Orr DP,
- Dittus RS
- 16.↵Johns Hopkins Hospital. The Harriet Lane Handbook. 18th ed. Philadelphia, PA: Mosby; 2008.
- Bachur RG,
- Hennelly K,
- Callahan MJ,
- Chen C,
- Monuteaux MC
- Copyright © 2013 by the American Academy of Pediatrics