We observed small-area variation in computed tomography scan use for inpatients in New York State, even after controlling for relevant patient and hospital characteristics.
To examine patient, hospital, and geographic characteristics influencing variation in computed tomography (CT) scan use in inpatients in New York State.
Retrospective cohort study.
We used the 2007 Healthcare Cost and Utilization Project’s State Inpatient Database from the Agency for Healthcare Research and Quality and applied descriptive univariate statistics and logistic regression models to quantify the influence of each factor on CT scan use.
The primary contributors to variation in CT scan use were the inpatients’ diagnosis, age, and hospital county, whereas inpatients’ sex and method of payment and hospitals’ teaching status and size had very little effect. Inpatients diagnosed with trauma had the highest CT scan use; CT scan use increased with age for inpatients over 30 years; and CT scan use varied widely between counties.
After controlling for patient and hospital characteristics, significant geographic variation remained at the level of the county, which indicates that additional research investigating the use of CT scans is necessary to understand the reasons behind small-area variation. Understanding the distribution and practice patterns of specific physician specialties may be helpful in curtailing underuse and overuse.
Am J Manag Care. 2013;19(3):e93-e99An inpatient’s diagnosis and age were the strongest predictors of the likelihood of receiving a computed tomography (CT) scan. The third strongest predictor among 8 patient, hospital, and geographic characteristics was the county in which the scan was administered.
Since computed tomography (CT) was first introduced in the 1970s, its use has grown to an estimated 72 million CT scans performed annually in the United States.1 Computed tomography scans are currently a fundamental diagnostic tool in the evaluation of a multitude of conditions, including malignancies, cardiovascular disease, and infectious diseases, among others. The largest increase in CT scan use in recent years has been for procedures such as virtual colonoscopy, cardiac screening, and whole-body CT scans for asymptomatic patients as well as smokers and children.2 With the increased use of CT scans in low-risk populations, some have begun to question the public health implications of unnecessary radiation exposure of patients.3-5
The ubiquity of CT scanners provides access to accurate data to diagnose complex medical problems. However, the availability of CT scanners may have created a supply-induced demand, which may contribute to the variability in practice and CT scan use.6,7 Computed tomography scans improve diagnostic accuracy for certain conditions. However, in some instances, such as appendicitis, the benefit to patients’ health may not be justified.8,9
Almost 40 years ago, Wennberg and Gittelsohn10 observed that more care is not always better care. Numerous publications after this study observed variations in medical usage patterns (eg, a study on small-area variation in New York State for cesarean section11). Understanding smallarea variation in CT scan use can help identify underuse and overuse, both of which may be costly and negatively affect healthcare quality.4,12 We investigated how patient, hospital, and geographic characteristics influence CT scan use for inpatients with the goal that our research will lead to more effective CT scan use, lower costs, and higher quality.
This research used the 2007 edition of the New York State Inpatient Database (SID) from the Healthcare Cost and Utilization Project. This data set resulted from a federal-state-industry partnership sponsored by the Agency for Healthcare Research and Quality. We analyzed 2,485,498 inpatients from 221 hospitals in 56 counties in New York State. In our analysis, we expunged fewer than 5% of the inpatient records in the SID because of missing values. No obvious patterns were evident in these expunged records compared with those that had complete data. As the data were de-identified, no institutional review board approval was requested or required.
The Agency for Healthcare Research and Quality created a utilization flag to indicate inpatient CT scan use based on International Statistical Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes, revenue codes, or both.13 In our study, an inpatient was considered to have received a CT scan if the inpatient had at least 1 report based on the utilization flag. We found that ICD-9-CM codes identified only 23.2% of inpatients receiving a CT scan, whereas revenue codes identified 98.8% of them.
We determined the association of CT scan use with the following categories of independent variables: (1) patient characteristics, (2) hospital characteristics, and (3) geographic characteristics. Patient demographics (including sex, race, and age), Major Diagnostic Category (MDC), payment method, and CT scan use were drawn directly from variables available in SID. For the age variable, we used small age group intervals to describe variation in CT scan use across age ranges. We followed conventional pediatric age groupings for ages <18 years (birth to 1 year, 1-4 years, 5-12 years, and 13- 17 years). The age groupings for adult inpatients were 18 to 29 years, 30 to 39 years, 40 to 64 years, 65 to 84 years, and >85 years. The MDCs divided patients into 25 mutually exclusive diagnostic areas based on the principal diagnosis from the ICD-9-CM codes.14
Hospital characteristics including hospital teaching status, location (urban or rural), and bed size were obtained from the New York State Department of Health.15 We used the Health Cost and Utilization Project’s classification to categorize hospitals as small, medium, or large based on their number of beds, location, and teaching status.16 The hospital’s county was the geographic characteristic we considered for this study.
To investigate the association of the independent variables with CT scan use, we calculated univariate statistics and reported mean values with their odds ratios (ORs)
and confidence intervals (CIs). We used a logistic regression model to predict the odds of an inpatient receiving a CT scan based on the independent variables discussed above. The urban or rural location of an inpatient’s hospital was not included in the logistic regression model because this variable was too highly collinear with the other variables in the model. We calculated the Schwarz criterion to determine the relative influence of each independent variable on explaining variation in CT scan use. The Schwarz criterion was calculated for the full model and separate submodels to evaluate the influence of each independent variable. For each submodel, we eliminated 1 independent variable and determined the difference in Schwarz criterion values between this submodel and the full model. The greater the difference between the Schwarz criterion values, the greater was the explanatory power of the independent variable.
RESULTSCT Scan Use by Inpatients
shows how the percentage of inpatients who received at least 1 CT scan deviated from the state average for each county. Across the state, 28.0% of inpatients received at least 1 CT scan. Cortland County had the highest rate with 37.4%, closely followed by Rockland County with 37.2%. As shown in Figure 1, these 2 counties had CT scan use of +9.4% and +9.2% over the state average, respectively. In contrast, Otsego County and Schuyler County had the lowest rates of CT scan use and the largest deviation magnitude. Their rates were 17.0% and 17.3%, respectively, which translates into deviations from the state average of -11.0% and -10.7%.
Overall, we found considerable geographic variation. The median absolute deviation from the state average was 3.6%. No correlation existed between these CT scan use percentages and county population, even when adjusted for numbers of inpatients (r = -0.01, n =56 counties, 95% CI, -0.27 to 0.26). Also, no correlation existed between CT scan use percentages and number of hospitals per county (r = -0.008, 95% CI, -0.27 to 0.26).
Patient Characteristics Associated With Inpatient CT Scan Use
Patient Demographics and CT Scan Use. shows inpatient CT scan use according to various patient characteristics. In this univariate analysis, there was statistically significant variation in inpatient CT scan based on sex, race, payer, age, and diagnosis. Males were more likely to have CT scans performed than females, with an OR of 1.172 (95% CI, 1.165-1.178). White inpatients were more likely to have CT scans than other racial groups. The ORs for racial groups compared with whites ranged from 0.687 (95% CI, 0.675- 0.697) for Asians to 0.881 (95% CI, 0.875-0.888) for blacks. Medicare benefi ciaries had the highest likelihood of receiving a CT scan, with an OR of 2.237 (95% CI, 2.222-2.252) compared with inpatients with private insurance.
Inpatient Age and CT Use Rate. We found significant variation in CT scan use for different ages. We calculated the percentage of inpatients of a given age who received a CT scan (). Inpatients aged 88 years had the highest rate of CT scan use, at 47.2% of 20,783 inpatients aged 88 years. Only 1.7% of inpatients younger than 1 year received a CT scan (n = 260,503). Use of CT scans increased by age from birth to 8 years (23.3%, n = 4082), stayed relatively fl at until age 15 years (23.7%, n = 8077), decreased until age 30 years (14.7%, n = 27,604), and then steadily increased until age 88 years. The percentages of CT scan use for the oldest inpatients in the data set were slightly lower than the percentage for the peak age of 88.
Inpatient Medical Condition and CT Use Rate. shows the association of MDCs with CT scan use. As expected, there was considerable variability in CT scan use across diagnostic categories. Trauma was the diagnostic category with the highest CT scan use (89.8%, n = 3377), while Newborn and Other Neonates (Perinatal Period) was the diagnostic category with the lowest use (1.0%, n = 234,290).
Hospital Characteristics Associated With Inpatient CT Scan Use
lists the CT scan use rates for differences in hospitals’ teaching status, location, and bed size. The inpatient CT scan use was 30.7% for nonteaching and 27.1% for teaching hospitals. The inpatient CT scan use was 28.1% for urban and 26.3% for rural hospitals. Only a weak positive correlation could be observed between CT scan use and hospital bed size (r = 0.15, n = 221 hospitals, 95% CI, 0.02-0.27).
Logistic Regression Results
To control for patient and hospital characteristics and determine residual geographic variation in CT scan use at the county level, we used a logistic regression model. Figure 3 shows how the percentage of inpatients who received at least 1 CT scan deviated from the model’s prediction for each county. Based on its patient and hospital characteristics, Schoharie County had the highest predicted CT scan use, 39.7%. Its actual use was 30.1%, showing a deviation of -9.6%. Jefferson County had the lowest predicted CT scan use, 21.5%. Its actual use was 20.7%, showing a deviation of -0.8%. Delaware County had the largest magnitude deviation. The model predicted that 33.9% of its inpatients would receive at least 1 CT scan, but only 20.7% of its inpatients did, for a deviation of -13.2%. Rockland County had the largest positive deviantion, +10.1%, from its model prediction. Its prediction was 27.2%, but 37.2% of its inpatients received at least 1 CT scan. lists for each county the observed percentage of inpatients receiving a CT scan, the expected percentage, the deviation, and the standardized ratio of observed percentage to expected percentage.
After controlling for patient and hospital characteristics, we still found considerable variation between counties. The median absolute deviation from the state average was 3.1%. Before using the logistic regression model (see Figure 1), the median absolute deviation was 3.6%.
The results of a logistic regression analysis that included county as an independent variable, along with the patient and hospital characteristics, can be found in . After controlling for the other factors, the results show that males, whites, and Medicare beneficiaries were not more likely to receive a CT scan, contrary to the results of the univariate analysis above. The OR for males compared with females was 0.977 (95% CI, 0.970-0.983). Compared with whites, the OR for Asians was 1.145 (95% CI, 1.123-1.168) and the OR for blacks was 1.128 (95% CI, 1.118-1.139). Medicare beneficiaries had an OR of 1.052 (95% CI, 1.041-1.068) compared with inpatients paying with private insurance. Inpatients who self-paid (OR = 1.228; 95% CI, 1.209-1.248) or were not charged (OR= 1.619; 95% CI, 1.505-1.743) were the most likely to have received a CT scan. An inpatient’s odds of receiving a CT scan generally increased with age. Compared with the reference group of 30- to 39-year-olds, the youngest inpatients (birth to 1 year) had the lowest odds of receiving a CT scan (OR = 0.109; 95% CI, 0.104-0.115), and the oldest patients (85 years and older) had the highest odds (OR = 1.604; 95% CI, 1.576-1.633). Trauma was the diagnostic category most likely to result in a CT scan (OR compared with the reference category of Endocrine, Nutritional and Metabolic System was 24.531; 95% CI, 21.888-27.493), while Pregnancy, Childbirth and Puerperium was the least likely category (OR = 0.049; 95% CI, 0.047-0.051). Inpatients at nonteaching hospitals were slightly less likely to receive a CT scan than those at teaching hospitals (OR = 0.971; 95% CI, 0.962-0.981). Inpatients at large hospitals (OR = 0.975; 95% CI, 0.967-0.984) or small hospitals (OR = 0.842; 95% CI, 0.832-0.852) were less likely to receive a CT scan than those at medium-sized hospitals.
We determined how efficient and effective each variable was in explaining CT scan use rates by using the Schwarz criterion. The presents Schwarz criterion values for the full model and the submodels. The greater the difference in Schwarz criterion values (Δ Schwarz criterion) between the full model and the submodel, the greater the efficiency and effectiveness of the corresponding categorical factor of the submodel in describing CT scan use rates. Results indicate that MDC was the variable associated with the greatest variation in CT scan use, followed by age group and county. Race, hospital bed size, and method of payment had little explanatory value, and sex and hospital teaching status played virtually no role in predicting CT scan use.
Geographic analysis can be a helpful tool for understanding care patterns, technology use, and small-area variations in healthcare. Several broad-based studies have demonstrated differences in health outcomes on a geographic level.10,11,17-20 However, the use of geographic analysis to determine the variability in radiology testing, specifically the use of CT scans, has not been performed. Therefore, we conducted a geographic analysis of CT scan use and examined how patient, hospital, and geographic factors affect the use of this technology for inpatients in New York State. We found signifi cant variations due to patient characteristics (age, race, insurance status, sex, MDC). In addition, we found that hospital characteristics (bed size, hospital teaching status) had a weak infl uence on CT scan use. Finally, after controlling for patient and hospital factors, unexplained variability remained at the level of the county. Identifying additional factors that may explain the residual variability is important for developing interventions and managing care to ensure the appropriate use of this important technology.
Since Wennberg and Gittelsohn10 fi rst developed their conceptual approach to understanding the effect of geographic variation on the use of medical services for surgical procedures, others have advanced their work both in content and methodology for a broad range of procedures and populations.10,11,17-20 One of the major fi ndings of these studies was that procedural variation is greatly infl uenced by patient and hospital factors. Without controlling for these factors, a geographic analysis on medical procedures could be biased. Our study also found several important patient and population level characteristics—including age, insurance status, and diagnosis—that influenced the geographic variation of CT scans. Although certain characteristics such as age or diagnosis cannot be modifi ed, interventions informed by evidencebased guidelines or protocols can help to enforce appropriate CT scan use and can ultimately reduce variation and improve healthcare value.
Teaching hospitals perform important societal functions including the provision of direct care for patients and the education of the next generation of healthcare professionals. Surprisingly, we found little difference in CT scan use rates between teaching and nonteaching hospitals. This finding was interesting given that teaching hospitals have medical trainees who are less experienced in identifying the appropriate indications for ordering these diagnostic tests. In addition, teaching hospitals often serve a population who have worse access to healthcare, more comorbid conditions, and fewer resources to care for them. Hence, overuse of certain technologies and therapies was expected. Since teaching hospitals are involved in the education of medical trainees and evidencebased medicine principles have become foundational in this setting, we speculate that our observation is related to better adherence to protocols and guidelines for CT scan prescribing. If true, generalizing these findings outside of teaching institutions to a broad geographic catchment area may help to address the important issues of guideline adherence and reduce variability.
Finally, after controlling for patient and hospital characteristics, we found that geographic variation persisted at the level of the county. This remaining variation may be related to the number of CT scanners available for use,21 the number of radiologists and other physicians in the region, differences in provider practice preferences,22 or the economic status of a county. Regulatory mechanisms such as the Certificate of Need program are in place in New York State to assist with the appropriate allocation of expensive technologies. The major drivers of CT scan use are prescribing physicians. Hence, variability in the geographic distribution of physicians generally and in specific specialties like radiology may be as responsible for CT scan use patterns as the presence of CT scanners. Practice pattern differences can be ameliorated through continuing education and peer review programs. Unfortunately, the relationship between these factors and CT use cannot be understood by using the existing data elements in SID, but can certainly be investigated using other data sources and methods.
Several limitations in our study must be considered. The SID includes only inpatient data and does not include CT scans performed on outpatients even if performed at the same hospital. This limitation does not affect the conclusions for inpatient CT scan use, but may influence conclusions at thepopulation level. In addition, while we used literature-defined approaches for identifying the performance of CT scans during inpatient stays in the SID, misclassification biases associated with poor documentation may exist. We believe that we have overcome this limitation to the extent possible by using complementary approaches for acquiring CT scan use rates. Finally, we included all factors potentially affecting CT scan use that were available through the data set. However, we are humbled by the number of other potential explanatory variables that may be operational at the level of the county but that are not contained in the SID.
Despite these limitations, we believe that this study has significant strengths because it demonstrates that area variation exists for an expensive medical imaging technology like CT. In addition, the relationship between CT scan use and age is an important finding and can serve as guidance to alter physician practices, particularly for those patients who may be harmed by CT scan overuse like children and adolescents, in whom lifelong radiation dosage may adversely affect health.3-5 Further, as acute care hospitals and physicians consider methods for integrating inpatient and outpatient care for the benefit of patient populations, analyses that incorporate geography are likely to provide practical insights that ensure the continuity of service while avoiding oversupply of an expensive resource. These results can help to inform Certificate of Need programs both inside and outside of New York State for regulating CT scan deployment. Finally, we believe our study provides guidance for future research investigating racial and ethnic disparities in the provision of healthcare services because it identifies other factors that may account for county-level geographic variation in CT scan use or the use of other technologies.
We gratefully acknowledge assistance from Virginia Tech’s Laboratory for Interdisciplinary Statistical Analysis.
Author Affiliations: From Department of Statistics (EAV), Grado Department of Industrial and Systems Engineering (XX, AH, CW, ADS), Virginia Tech, Blacksburg, VA; Barnabas Health (ADS), West Orange, NJ.
Funding Source: This research was supported by the 2009 Research Acceleration Program at Carilion Clinic.
Author Disclosures: The authors (EAV, XX, AH, CW, ADS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (EAV, AH, CW, ADS); acquisition of data (CW); analysis and interpretation of data (EAV, XX, AH, CW, ADS); drafting of the manuscript (EAV, XX, AH, CW, ADS); critical revision of the manuscript for important intellectual content (EAV, XX, CW, ADS); statistical analysis (EAV, XX); provision of study materials or patients (CW); obtaining funding (EAV, AH, CW, ADS); administrative, technical, or logistic support (ADS); and supervision (EAV, CW, ADS).
Address correspondence to: Eric A. Vance, PhD, 212 Hutcheson Hall (0439), Blacksburg, VA 24061. E-mail: firstname.lastname@example.org. Mettler FA Jr, Thomadsen BR, Bhargavan M, et al. Medical radiation exposure in the US in 2006: preliminary results. Health Phys. 2008; 95(5):502-507.
2. Brenner DJ, Hall EJ. Computed tomography—an increasing source of radiation exposure. N Engl J Med. 2007;357(22):2277-2284.
3. Berrington de González A, Mahesh M, Kim KP, et al. Projected cancer risks from computed tomographic scans performed in the United States in 2007. Arch Intern Med. 2009;169(22):2071-2077.
4. Smith-Bindman R, Lipson J, Marcus R, et al. Radiation dose associated with common computed tomography examinations and the associated lifetime attributable risk of cancer. Arch Intern Med. 2009; 169(22):2078-2086.
5. Redberg RF. Cancer risks and radiation exposure from computedtomographic scans: how can we be sure that the benefits outweigh the
risks? Arch Intern Med. 2009;169(22):2049-2050.
6. Wu SR, Shakibai S, McGahan JP, Richards JR. Combined head and abdominal computed tomography for blunt trauma: which patients with minor head trauma benefit most? Emerg Radiol. 2006;13(2):61-67.
7. Lysdahl KB, Hofmann BM. What causes increasing and unnecessary use of radiological investigations? a survey of radiologists’ perceptions. BMC Health Serv Res. 2009;9:155.
8. Lee SL, Walsh AJ, Ho HS. Computed tomography and ultrasonography do not improve and may delay the diagnosis and treatment of acute appendicitis. Arch Surg. 2001;136(5):556-562.
9. Partrick DA, Janik JE, Janik JS, Bensard DD, Karrer FM. Increased CT scan utilization does not improve the diagnostic accuracy of appendicitis in children. J Pediatr Surg. 2003;38(5):659-662.
10. Wennberg J, Gittelsohn A. Small area variations in health care delivery. Science. 1973;182(4117):1102-1108.
11. Epstein AJ, Nicholson S. The Formation and evolution of physician treatment styles: an application to cesarean sections. J Health Econ. 2009;28(6):1126-1140.
12. Brenner D, Elliston CD, Hall E, Berdon W. Estimated risks of radiation- induced fatal cancer from pediatric CT. AJR Am J Roentgenol. 2001;176(2):289-296. 13. Agency for Healthcare Research and Quality. Utilization flags. http://www.hcup-us.ahrq.gov/toolssoftware/util_flags/utilflag.jsp. Published November 2010. Accessed August 4, 2012.
14. Utah Department of Health. Major Diagnostic Categories (MDC). http://health.utah.gov/opha/IBIShelp/codes/MDC.htm. Updated September 2005. Accessed July 13, 2012.
15. New York State Department of Health. New York State hospital profile. http://hospitals.nyhealth.gov/. Published 2011. Accessed November 28, 2011.
16. Healthcare Cost and Utilization Project. NIS description of data elements: bedsize of hospital. http://www.hcup-us.ahrq.gov/db/vars/ hosp_bedsize/nisnote.jsp. Published September 2008. Accessed November 28, 2011.
17. Chassin MR, Kosecoff J, Park RE, et al. Does inappropriate use explain geographic variations in the use of health care services? a study of three procedures. JAMA. 1987;258(18):2533-2537.
18. Volinn E, Diehr P, Ciol MA, Loeser JD. Why does geographic variation in health care practice matter? (and seven questions to ask in evaluating studies on geographic variation). Spine (Phila Pa 1976). 1994;19(18)(suppl):2092S-2100S.
19. Rao VM, Parker L, Levin DC, Sunshine J, Bushee G. Use trends and geographic variation in neuroimaging: nationwide Medicare data for 1993 and 1998. AJNR Am J Neuroradiol. 2001;22(9):1643-1649.
20. Havranek EP, Wolfe P, Masoudi FA, Rathore SS, Krumholz HM, Ordin DL. Provider and hospital characteristics associated with geographic variation in the evaluation and management of elderly patients with heart failure. Arch Intern Med. 2004;164(11):1186-1191.
21. Niagara Health Quality Coalition. MRI Services in Western New York 2003. http://www.myhealthfinder.com/hcac/MRIreport03.pdf. Published February 2004. Accessed November 28, 2011.
22. Gawande A. The cost conundrum: what a Texas town can teach us about health care [Annals of Medicine]. The New Yorker. http://www .newyorker.com/reporting/2009/06/01/090601fa_fact_gawande. Published June 1, 2009. Accessed November 28, 2011.