To systematically review algorithms identifying cases of pancreatitis in administrative data, with a focus on studies examining algorithm validity.
To systematically review algorithms identifying cases of pancreatitis in administrative data, with a focus on studies examining algorithm validity.
A literature search was conducted using PubMed and the Iowa Drug Information Service database. Reviews were conducted by two investigators identifying studies using data sources from the USA or Canada. These data sources most likely reflect the coding practices of Mini-Sentinel data partners.
Eight studies were obtained examining the validity of an algorithm to identify pancreatitis in either hospital or ambulatory medical records or billing databases. The best-performing algorithm was International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 577.0; with a positive predictive value of 60%–80% and a negative predictive value usually greater than 90%. Populations involved in different studies were heterogeneous; age ranges, level of population risk, pancreatitis etiology, and geographic distribution were highly variable.
Validation studies find that the principal ICD-9-CM diagnosis code of 577.0 had the best positive predictive value and specificity. Current studies do not support the use of the ICD-9-CM codes 577.1 and 577.2. Databases enhanced with laboratory values at point of care would invariably increase the specificity of existing algorithms. Copyright © 2012 John Wiley & Sons, Ltd.
The US Food and Drug Administration (FDA) commissioned systematic reviews to identify validation studies of algorithms to identify 20 health outcomes of interest (HOI) in administrative and claims data, as part of its Mini-Sentinel pilot program. These reviews provide the foundation for future studies of HOIs in Mini-Sentinel and other administrative data sources. In such studies, it is extremely important to understand the performance characteristics of the codes that might be used to identify an HOI, as the presence of a code is not always sufficient to determine that an HOI actually occurred.
Pancreatitis is an inflammatory process of the pancreas that can be acute or chronic requiring treatment ranging from minor outpatient management up to intensive care from multiorgan failure. Thus, depending on the severity of disease, the corresponding mortality rates vary from 2% to 30%. In the USA alone, pancreatitis accounts for over 210, 000 hospitalizations each year. Etiologies of pancreatitis vary, with gallstones accounting for the majority of all cases, followed by alcohol. Other common causes of pancreatitis include trauma, ischemia, mechanical obstruction, infections, autoimmune, and rarely, drugs. Notably, up to 100 drugs have been implicated in causing acute pancreatitis, yet many lack robust evidence supporting a causal link.
The current International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) terminology code for acute pancreatitis (577.0) entered circulation in 1977. This code refers to “acute or chronic inflammation of the pancreas due to auto-digestion of the pancreatic tissue by its own enzymes” and “inflammation of the pancreas with pain as the primary symptom.” This definition also covers other associations or etiologies of the pancreatic disease; some inflammatory (e.g., calcareous, gangrenous, acute interstitial, malignant, suppurative, and subacute) and some noninflammatory (e.g., hemorrhagic, annular, apoplectic). The major limitation of the current 577.0 terminology is that it does not allow for etiological or a pathophysiological description of the kind of acute pancreatitis.
The clinical diagnostic criteria for the assessment of acute pancreatitis have remained stable for the past 3 decades, largely relying on three components: history, laboratory, and radiology examination. The key historical elements are (i) history of acute onset of mid-epigastric pain, (ii) with radiation to the back, (iii) combined with the presence of either nausea or vomiting, and (iv) aggravation by oral consumption. Key laboratory elements include elevations (more than three times the normal) of amylase, lipase, or both. Finally, in cases where the clinical history or laboratory examination is uncertain, computed tomography (CT) of the abdomen revealing pancreatic inflammation can secure the diagnosis.
This article provides an overview of the pancreatitis review. The full report can be found at http://mini-sentinel.org/foundational_activities/related_projects/default.aspx.
Details of the methods for these systematic reviews can be found in the accompanying article by Carnahan and Moores. In brief, the base PubMed search was combined with the following terms to represent the HOI: “pancreatitis” [All Fields] OR “pancreatitis/chemically induced” [All Fields] OR “pancreatitis/diagnosis” [All Fields] OR “pancreatitis/epidemiology” [All Fields] OR “pancreatitis acute necrotizing” [All Fields] OR “pancreatitis acute” [All Fields] OR “pancreatitis, acute” [All Fields] OR “pancreatitis, acute necrotizing” [All Fields] OR “pancreatitis, acute necrotizing/diagnosis” [All Fields] OR “pancreatitis, acute necrotizing/epidemiology” [All Fields] OR “pancreatitis, acute necrotizing/etiology” which was combined with “pancreatitis” [Mesh] OR “pancreatitis, acute necrotizing” [Mesh]. Searches of Embase and the citation database of the Iowa Drug Information Service (IDIS) were also conducted. The HOI in the IDIS database was searched with two methods, one in which the descriptor SIDE EF DIGESTIVE 78 was used and another in which the Disease Term 577.* was used. Note that because the IDIS database uses the ICD-9-CM codes as disease terms, a truncation of that term was used. The details of these searches can be found in the full report on the Mini-Sentinel Web site. The PubMed and the IDIS searches were conducted on 23 June 2010. All searches were restricted to articles published in 1990 or after. Mini-Sentinel collaborators were also asked to help identify any relevant validation studies.
The abstract of each citation identified was reviewed by two investigators. When either investigator selected an article for full-text review, the full text was reviewed by both investigators. Agreement on whether to review the full text or include the article in the evidence table was calculated using a Cohen's kappa statistic. In instances where there were fewer than five studies validating a specific algorithm, algorithms without validation are reported. The data in the evidence table were extracted by one and confirmed by a second investigator.
Of the 158 abstracts reviewed, 62 were selected for full-text review; 40 were excluded because they did not study the HOI, 42 were excluded because they were not administrative database studies, and 14 were excluded because the data source was not from the USA or Canada. Cohen's kappa for agreement between reviewers on inclusion versus exclusion of abstracts was 0.60. The major source of disagreement was whether a study was an administrative database study. This related to the decision to review single center studies that may have used coding algorithms and the difficulty in discerning the likelihood of a coding algorithm when reviewing abstracts of such studies.
Of the 62 full-text articles reviewed, 6 were included in the final evidence table of validation studies; 23 were excluded because the HOI identification algorithm was poorly defined, and 21 were excluded because they included no validation of the outcome definition or reporting of validity statistics. Reviewers identified two citations for review from full-text article references. Both were included in the final report. Cohen's kappa for agreement between reviewers on inclusion versus exclusion of full-text articles reviewed was 0.77. The most common reason for disagreement on inclusion of a study was the lack of validation of the outcome definition or reporting of validity statistics which occurred in two studies. Agreement was reached by discussion.
Mini-Sentinel collaborators provided no validation studies for inclusion in the report. One additional study with a validated algorithm has been published since the time the original search was done, and that study has been added in the summary of algorithms below and in Table 1 of algorithm validation studies.
|References||Study population and time period||Description of outcome studied||Algorithm||Validation and adjudication procedure and operational definition and validation statistics|
|Yadav et al.||Male veterans attending the outpatient detoxification program at the Central Arkansas Veterans Healthcare System. January 2002 to December 2003. A total of 1409 patients were treated during this period. Patients with gallstones were excluded. All subjects in the sample had an ICD-9 diagnosis code for alcohol abuse. The mean age was 47.6 years, 43.6% were White, 90% had drug use at some time, 90% were current smokers, and 27% had hepatitis C.||AP or chronic pancreatitis||ICD-9 code 577.0 for AP or ICD-9 code 577.1 for chronic pancreatitis||The medical record for 87 patients with an ICD-9 code for pancreatitis and a random sample of 214 subjects without the ICD-9 code were reviewed by a study gastroenterologist. Confirmation of AP was based on the presence of typical abdominal pain with elevation of amylase, lipase, or both 3 times or more than normal and/or imaging evidence of pancreatitis. Alcoholic chronic pancreatitis was confirmed based on (i) typical history of recurrent episodes of AP, except primary painless chronic pancreatitis; (ii) history of excessive alcohol intake ≥80 g/day for years plus one of the following: (a) calcification in the pancreas (x-ray or CT), (b) moderate or marked changes in ductal system on ERCP, (c) marked exocrine insufficiency, and (d) typical histology on adequate surgical specimen.|
|Chart review verified 40 of the 87 patients with an ICD-9 code for pancreatitis: PPV = 46%.|
|For patients with both codes 577.0 and 577.1, the PPV was 77%; for code 577.0 only, it was 40%; and for code 577.1 only, it was 20%.|
|Of 214 patients without an ICD-9 code for pancreatitis, 2 had chronic pancreatitis confirmed by chart review: NPV = 99%.|
|Guo et al.||Ohio Medicaid patients with a diagnosis of HIV and at least one prescription for an antiretroviral medication. n = 4972 January 1997 to December 2002. Mean age 36.1 ± 10.7 years, 27% were women, 48% were Caucasian, 44% were African American, 10% had diabetes, and 44% had a preexisting mental illness.||AP||ICD-9 code 577.0 and at least one laboratory order for either amylase or lipase indicated by CPT codes 82150 or 83690 within 30 days before or after the date of diagnosis||Random sample of 20 patients with AP had their inpatient and outpatient medical claims records reviewed. The record was examined for at least one sign or symptom of nausea, vomiting, abdominal pain, or tenderness accompanied by elevation of serum amylase or lipase level. ICD-9 787.03 for nausea or vomiting, 789.0 for abdominal pain, and 789.60–789.69 for abdominal tenderness. They also expected discontinuation of one or both nucleoside reverse transcriptase inhibitors (NRTIs) or protease inhibitors or a switch in antiretroviral drugs.|
|Excluded patients ≥65 years||The validation process was described in the methods, but no results were reported. The lead author was contacted to request information on these results, but none was received.|
|McMenamin and Gates||All patients from the University of Kentucky Chandler Medical Center in 1992 with discharge ICD-9 code 577.0, n = 115. Mean age 42 years, 56% men, etiology of pancreatitis: alcohol, 17; unknown, 21; ERCP, 9; drug, 3; stone, 3; hyperlipidemia, 2; and malignancy, 2. Six deaths occurred.||AP||ICD-9 code 577.0 in discharge records||Medical records were available for 95 of the 115 cases and were reviewed for clinical evidence of pancreatitis. Clinical evidence included abdominal pain, with an assessment noted of pancreatitis in the progress notes, serum amylase >2 times the upper limit of normal, and possibly an inability to tolerate a normal diet, but this is not stated clearly. These were the factors used in the study to document duration of the episode of AP.|
|22 of 95 cases had a lack of documented clinical evidence of pancreatitis: 73/95 PPV= 76.8%.|
|Morton et al.||Baseline data from 1978 to 1985 of 128 934 adults in the Kaiser Permanente Medical Care in San Francisco and Oakland. Records screened to 31 December 1998. 50% were men, 48.5% were <50 years old, 35 % were Black, 51% were white or Hispanic, and 7.5% were Asian.||AP or chronic pancreatitis||Hospitalization for possible pancreatitis was screened for a primary discharge diagnosis with code 577.||A medical record analyst performed initial chart screening, with a physician investigator reviewing all final diagnoses. A diagnosis of pancreatitis was considered confirmed if symptoms and physical examination were compatible and the serum amylase was elevated. In cases without elevated serum amylase, evidence of pancreatitis by imaging or direct examination of the organ was required.|
|452 persons were initially screened, and 439 were confirmed to have pancreatitis. PPV was not reported by the authors, but it is 97% based on the data reported. This study did not distinguish between acute or chronic pancreatitis.|
|Park et al.||Yale-New Haven Children's Hospital, New Haven, CT tertiary care teaching hospital with a broad catchment area. Patients' age, birth to 21 years, seen between August 1994 and July 2007. Mean age was 13.1 ± 5.64 years, 40% were males, 53% were White, 23% were Black, and 19% were Hispanic. The most common causes of AP were biliary 32.6%, medications 25.6%, idiopathic 20%, systemic 10%, trauma 9%, viral infection 8%,5% metabolic condition, and 4% ERCP.||AP||ICD-9 code for AP, not stated but assumed to be 577.0||Records were reviewed for inclusion criteria relevant to AP. To be included in the study group, patients needed any 1 of the following 3 features:|
|(i) Serum amylase or lipase greater than 3 times the upper limit of normal. (ii) Radiographic evidence of AP on computed tomography and ultrasound demonstrating a minimum of pancreatic parenchymal changes or peripancreatic fluid. (iii) Serum lipase greater than 1.5 times the upper limit of normal that could not be explained by nonpancreatic causes and the presence of 2 out of 3 clinical features—abdominal pain characteristic of AP, nausea and vomiting, or epigastric tenderness|
|282 patients were identified by ICD-9 codes. 215 met the inclusion criteria of a confirmed case of pancreatitis. 18 were excluded as chronic pancreatitis. The others had incomplete data. 215/282 is a PPV of 76.2%. PPV was not calculated by the authors.|
|Quraishi et al.||Patients in the Henry Ford Health System in Detroit who were initiated on dialysis after 1 January 1998. Data were collected retrospectively from 1 January 1998 to 1 August 2003.||AP||ICD-9 codes for pancreatitis and pancreatitis-related complications in the admitting and discharge diagnoses. The following ICD-9 codes were used: 577.0, 577.1, and 577.2.||AP is defined by the presence of abdominal pain in association with elevation of serum amylase, lipase, or both >3 times the upper limit of normal and absence of another syndrome that could produce a similar presentation. Electronic medical records and paper charts were reviewed.|
|128 patients were identified by ICD-9 codes for AP and AP-related complications. No information about abdominal symptoms or elevated enzyme levels were found for 80 patients, 14 had cholecystitis with no evidence of AP, and 6 had diabetic ketoacidosis with no evidence of AP. 28 met criteria for AP; a PPV of 22%. 107 patients without ICD-9 codes for pancreatitis but with elevated amylase or lipase were found in electronic records. Manual chart review found that none met criteria for AP: NPV= 100%. The authors did not report specific performance statistics of the algorithm. It is unclear if the use of multiple ICD-9 codes had been included to produce high sensitivity at the expense of specificity of the algorithm. That appears to be true, but the authors did not address this issue or report which ICD-9 codes were most commonly reported.|
|Dore et al.||Initiators of antihyperglycemic drugs between 1 June 2005 and 31 December 2007, in the Normative Health Information database. 50% were women; aged <40 years, 19.2%; 40–54 years, 37.8%; 55–64 years, 30.9%; 65+, 12.1%. Type 2 diabetes mellitus, 61.5%.||Validate ICD-9 code 577.0||ICD-9 code 577.0 for AP in the first position, or in any position, of an emergency department or hospitalization claim||PPV for 577.0 in any position was 50%.|
|PPV for 577.0 in first position was 60%.|
|PPV based on age in the sample with 577.0 in the first position ranged from 51% in the oldest group to 74% in the youngest group. Of the 156 medical records obtained of patients with a CPT code for amylase or lipase tests without an ICD-9 code for pancreatitis, none had confirmed AP: NPV = 100%.|
|Yadav and Dhir||All male veterans with any pancreatitis ICD-9 codes (577.0 or 577.1) at the Little Rock AR VA between January 2001 and December 2002||AP or chronic pancreatitis||ICD-9 code 577.0 for AP or ICD-9 code 577.1 for chronic pancreatitis||AP was verified in the medical records by the presence of typical abdominal pain with ≥3 times elevation in serum amylase or lipase or imaging evidence of pancreatitis. Chronic pancreatitis was verified per 1997 International Workshop on CP (Pancreas 1997; 14: 215–221).|
|Plus 300 random controls without pancreatitis codes in the same period||Verified pancreatitis was present in 63% of the 418 patients with a pancreatitis code. The negative predictive value of no pancreatitis code was 98%.|
|Saligram et al.||Patients (n = 391) who received a primary discharge diagnosis of AP (ICD-9 577.0) after admission through the emergency room in years 2000, 2002, and 2005 at two University of Pittsburgh Medical Center hospitals (university, n = 200, and community, n = 191). Each patient was matched to the closest control (n = 391),||AP||ICD-9 code 577.0||Medical records were reviewed for the presence of characteristic abdominal pain with ≥3 times elevation of amylase and/or lipase or imaging evidence of AP as the “gold standard” for AP diagnosis. Two traditional etiologies, that is, gallstones and alcohol were present in 34% and 14% of cases, respectively.|
|Median age, 54 years; 50% men; 84% Caucasian||For all cases of AP, the PPV was 77%, and an NPV in controls was determined to be 97%.|
All publications listed in the evidence table used an ICD-9-CM code to identify patients with pancreatitis (Table 1). The majority of studies used ICD-9-CM code 577.0 for acute pancreatitis. Some used both 577.0 and 577.1 chronic pancreatitis. One study also used 577.2, which represents cyst and pseudocyst of the pancreas. Use of that additional code most likely produced a low specificity without increasing sensitivity for either acute or chronic pancreatitis and would not be recommended. One study also used common procedural terminology (CPT) codes 82150 or 83690 for either serum levels of amylase or lipase and ICD-9-CM codes 787.03 for nausea or vomiting, 789.0 for abdominal pain, and 789.60–789.69 for abdominal tenderness. It may be helpful for further research to determine if these codes increase sensitivity of an algorithm; however, based on current evidence, it would most likely significantly decrease specificity without a substantial increase in sensitivity.
Validation of the algorithm was primarily done by review of medical records. Validation of pancreatitis was based on physician notes of the description of the patient's pain and the presence of an elevated serum amylase or elevated lipase. In some instances, the physician's note of a presumptive diagnosis of pancreatitis was sought, and in some cases, evidence of pancreatitis documented by imaging was considered. The most common imaging study was CT. These methods for validation are consistent with the standard diagnosis of pancreatitis.
The various studies collectively included a broad range of subject characteristics to include the important ranges of age and gender, both high and typical risk levels for pancreatitis, a full range of the most common etiologies of pancreatitis, prevalent and incident pancreatitis, and spanning a broad geographic distribution. Identification of the outcome was based on either hospital discharge data or outpatient billing records.
Yadav et al. studied a population of male veterans with an average age of 47.6 years, 43.6% were White, and all had heavy alcohol use. The prevalence of pancreatitis (either acute or chronic) was 3 % in this population; 93% of cases were believed to be caused by alcohol. The study identified acute pancreatitis (577.0) or chronic pancreatitis (577.1) on the basis of an inpatient or outpatient evaluation. Upon chart review, 40 of the 87 patients had a confirmed diagnosis of pancreatitis for a positive predictive value (PPV) of 46%. PPV increased to 77% if both ICD-9-CM codes were present in the medical record. If only the code for acute pancreatitis (577.0) was present in the patient's record, the PPV was 40%, and the PPV was only 20% if the record contained only the code for chronic pancreatitis (577.1). In separate analyses, from a sample of patient records without an ICD-9-CM code for pancreatitis (n = 1409), 214 random charts were examined demonstrating a negative predictive value of 99%. Thus, among an at-risk group, ICD-9-CM codes were very sensitive to identification of the disease but had somewhat limited specificity.
Guo et al. studied the incidence of acute pancreatitis among a population of Ohio Medicaid patients younger than 65 years being treated for human immunodeficiency virus (HIV) infection. They used inpatient and outpatient claims to identify algorithm of an ICD-9-CM code 577.0 combined with at least one laboratory order for either amylase or lipase indicated by CPT codes 82150 or 83690 within 30 days of the diagnosis. Mean age was 36.1 years, 27% were women, 48% were Caucasian, and 44% were African American. Comorbid medical conditions included diabetes (10%) and preexisting mental illness (44%). The cases of pancreatitis in patients with newly treated HIV were incident cases, and the incidence rate was 1.95 per 100 person-years. Among patients newly treated for HIV, the cumulative incidence of pancreatitis was 4.2% and 2.6% in patients whose HIV infection had been diagnosed previously. Unfortunately, although a validation procedure for the algorithm was described in the methods of this study, no results were reported. Multiple attempts were made to contact this author with no response.
In a single hospital setting, McMenamin and Gates studied consecutive admissions with a discharge diagnosis code for acute pancreatitis, ICD-9-CM code 577.0. The mean age was 42 years, 44% were women, and typical etiologies for acute pancreatitis were identified, for example, alcohol, endoscopic retrograde cholangiopancreatography (ERCP), drug, bile stones, hyperlipidemia, and malignancy. Medical records were available for 95 of the 115 cases and were reviewed for clinical evidence of pancreatitis. The results of the validation process demonstrated a PPV of 76.8%.
Morton et al. studied the first hospital episode of pancreatitis, with a primary discharge diagnosis of ICD-9-CM code 577.* in the Kaiser Permanente Medical Care Program in San Francisco. Out of 129 000 eligible patients, 439 subjects with pancreatitis were identified; 48.5% were <50 years old; 50% were men; 51% were White or Hispanic; 35% were Black; and 7.5% were Asian. Etiologies of the pancreatitis included cholelithiasis (38%), alcohol (29%), idiopathic (25%), and miscellaneous (8%). The total incidence rate was 0.28 per 1000 person-years. This study did not distinguish between acute or chronic pancreatitis, although only the first hospitalization for pancreatitis was considered. In addition, mild cases of pancreatitis may not be hospitalized and may have been unrecognized. The PPV was 97%, calculated based on the proportion of potential cases excluded due to no evidence of pancreatitis in their medical records.
Park et al. studied acute pancreatitis in children. The mean age was 13.1 ± 5.64 years, and 40% were males, 53% were White, 23% were Black, and 19% were Hispanic. The most common causes of acute pancreatitis were biliary (32.6%), medications (25.6%), idiopathic (20%), systemic (10%), trauma (9%), viral infection (8%), metabolic condition (5%), and ERCP (4%). This population was hospitalized at a tertiary referral center. The algorithm for acute pancreatitis was presumably an ICD-9 code of 577.0. A total of 282 patients were identified by ICD-9-CM codes; 215 met the criteria of a confirmed case of pancreatitis, whereas 18 were excluded as chronic pancreatitis. The others had incomplete data. Although not reported by the authors, 215 confirmed cases of pancreatitis in the sample of 282 represent a PPV of 76.2%. However, if the 18 with chronic pancreatitis were also confirmed, the PPV would be 82.6%.
Quraishi et al. studied patients in the Henry Ford Health System in Detroit who were initiated on dialysis after 1 January 1998. Using retrospective hospital discharge records data from 1998 to 2003, ICD-9-CM codes 577.0, 577.1, and 577.2 identified 128 patients with pancreatitis and pancreatitis-related complications. A medical chart review defined acute pancreatitis by the presence of abdominal pain with elevation of serum amylase, lipase, or both more than three times the upper limit of normal (ULN) and the absence of a competing abdominal pain differential. In 80 patients, medical charts lacked information regarding abdominal pain, 14 had cholecystitis with no evidence of acute pancreatitis, and 6 had diabetic ketoacidosis with no evidence of acute pancreatitis. Only 28 met criteria for acute pancreatitis, indicating a PPV of 22%. A total of 107 patients without ICD-9-CM codes for pancreatitis had elevated pancreatic enzymes in the medical chart, and upon further review, none of these patients met criteria for acute pancreatitis, representing a negative predictive value of 100%. It is unclear if the use of multiple ICD-9-CM codes, in particular, 577.2, had been included to produce high sensitivity at the expense of specificity of the algorithm.
Dore et al. estimated the PPV of claims for acute pancreatitis among initiators of antihyperglycemic drugs between 1 June 2005 and 31 December 2007 in the Normative Health Information (NHI) database. The NHI dataset is a commercial health insurance transaction database that records ICD-9 and CPT codes. Among 260 255 initiators of antihyperglycemic drugs, medical records were sought for 842 potential cases of acute pancreatitis with an ICD-9 code of 577.0 in any position of an emergency department or hospitalization claim. Two practicing gastroenterologists assessed each case of the 585 charts that were obtained to confirm the diagnosis of acute pancreatitis based on the presence of at least two of the following: abdominal pain; serum lipase ≥3 times the ULN or a value of 300 U/L or higher without a specified normal range or serum amylase ≥5 times ULN or a value of 1000 U/L or higher without a specified normal range; and magnetic resonance imaging or CT scan with interpretation of peripancreatic fluid collection or streaking, or pancreatic edema or necrosis. These investigators also abstracted a random sample of medical records of patients with CPT codes for amylase or lipase tests (CPT 82150 or 83690) associated with an inpatient or emergency department encounter, but without a diagnosis code for acute pancreatitis, to assess the sensitivity of the claims-based case screening. They found an overall PPV for the ICD-9 code 577.0 in any position to be 50% and in the first position to be 60%. They also found that the PPV varied by age group in the sample of 409 records with the ICD-9 code in the first position. For those younger than 40 years, the PPV was 74%; for 40–54 years, it was 63%; for 55–64 years, it was 57%; and for older than 65 years, it was 51%. However, the 95% confidence intervals for these PPVs showed obvious overlapping. Of the 156 medical records obtained of patients with a CPT code for amylase or lipase tests without an ICD-9 code for pancreatitis, none had confirmed acute pancreatitis, representing a negative predictive value of 100%.
Two validation studies have been published as abstracts only.[12, 13] One study involved a population of veterans in which ICD-9-CM codes 577.0 and 577.1 were validated. The other involved patients with a primary discharge diagnosis acute pancreatitis ICD-9-CM code 577.0 at two University of Pittsburgh Medical Center hospitals. The PPV in the Veterans Affairs study was approximately 65%, and it was approximately 77% in the University of Pittsburgh study. Both studies found a negative predictive value of approximately 97%.
A description of two nonvalidated algorithms is included in the evidence table (Table 2) because they demonstrate that the most frequently validated algorithm of the ICD-9-CM code of 577.0 has in one case been used in a very large inpatient care database by Brown et al. This report provides some potentially useful evidence of the increasing incidence of acute pancreatitis and other demographic characteristics of patients diagnosed with this disease. Dore et al. demonstrated the use of ICD-9-CM code of 577.0 in a large database to examine the incidence of pancreatitis associated with the use of selected medications for the treatment of diabetes.
|Reference||Study population and period||Description of outcome studied||Algorithm|
|Brown et al.||The National Inpatient Sample database is the largest all-payer inpatient care database in the USA. The database contains discharge data from 994 hospitals representing 37 states. The hospitals which provide data to the NIS represent a stratified sample of 20% of all US community hospitals. This study covered the period 1997–2003. Mean age, 52.6 years; men, 51.0%. During the study period, there were 1 476 498 admissions with a principal discharge diagnosis of acute pancreatitis. The frequency of AP discharges increased by 30.2% from 1997–2003. Most discharges with AP were in the southern USA and were at large nonteaching hospitals located in urban areas.||Acute pancreatitis||ICD-9 code 577.0 as the principle discharge diagnosis (acute pancreatitis)|
|Patient demographics show that 51% are males, 1.6% are aged 1–17 years, 35.6% 18–44 years, 34.8% 45–64 years, 23.3% 65–84 years, and 4.7% 85+ years. The mortality rate declined from approximately 1.9% in 1997 to 1.4% in 2003. The mean length of stay declined from 6.4 days to 5.8 over this period.|
|Dore et al.||Data were obtained from the proprietary Ingenix Research Datamart from June 2005 through June 2008. 27 996 initiators of exenatide, 44 264 initiators of metformin/glyburide, and 16 276 initiators of sitagliptin. Age range, from younger than 19 to older than 60 years, 40% were 50–59 years, approximately 50% were women.||Acute pancreatitis||Hospitalization claims with a primary ICD-9 diagnosis code of 577.0 (acute pancreatitis)|
Studies that included a primary hospital discharge diagnosis of ICD-9-CM code 577.0 appeared to have the best PPV and specificity. This code was also the one most commonly used. The major limitation of the current 577.0 terminology is that it does not allow for etiological or a pathophysiological description of the kind of acute pancreatitis; a clear difference between ICD-9-CM and ICD-10 coding algorithms. Consideration of this must be given weight because as of October 2013, physicians will be required to code using ICD-10 terminology. Studies validating acute pancreatitis using 577.0 indicate a range of PPV from 40% to75%.[5, 7] For the most part, these validation studies relied on chart abstraction methods and used screening personnel (e.g., research assistant) confirmed by an expert reviewer (e.g., physician or gastroenterologist). The information abstracted was consistent with clinical criteria, but in several cases, poor medical charting limited the studies' ability to find a higher PPV for the algorithm.
There was only one study attempting to validate chronic pancreatitis using 577.1 in which the PPV was 20%. Combining the lack of clinical agreement on diagnosing chronic pancreatitis with the lack of studies specifically examining the validity (or added sensitivity and specificity) of using 577.1, it is not recommended. A recommendation not to use 577.2 would be based on similar reasoning. Identifying incident cases of acute pancreatitis should be relatively straightforward as most cases of 577.0 will be recorded on a hospital discharge record. Screening for prior primary and secondary codes for 577.0 in a population over a period should be feasibly accomplished. The main difficulty will be identifying populations at risk for the development of acute pancreatitis.
Further study is warranted to determine how to use procedure codes as part of the algorithm to determine the presence of acute pancreatitis in the absence of 577.0 but with either 577.1 or 577.2. In such cases, the presence of a CPT code for amylase or lipase, or the receipt of an abdominal CT with peripancreatic inflammatory findings, could enhance both PPV and maintain specificity.
Due to the variability in performance characteristics identified and single center nature of most studies, it may be useful to perform validation studies that use case finding algorithms from large administrative databases that include a broad population base. In particular, focusing on databases that are enhanced with actual laboratory data gathered at point of care could lend additional confidence in selecting the best algorithm. Performance characteristics seemed to vary, somewhat unpredictably, among studies of groups with different baseline risks of pancreatitis. This is a notable limitation in the current studies because most populations were defined by a unique illness (e.g., HIV or end-stage renal disease) or to be contained within a specific healthcare system. Finally, algorithms have not been validated in the very elderly, and validation studies have not been performed with ICD-10 codes.
Studies that included a primary hospital discharge diagnosis of ICD-9-CM code 577.0 had the best PPV and specificity. A PPV of 60%–80% may be reasonably obtainable with this algorithm, although in some instances, it was as low as 40%. The negative predictive value is expected to be 90% or better. This algorithm has been evaluated in a wide range of individual populations, with some involving high-risk individuals (heavy alcohol use, dialysis, HIV, gallstones) and others with typical risk populations. Some studies involved pediatric patients, and other studies included the primary age range of this condition (i.e., 45–55 years). The studies have included individuals with various etiologies for pancreatitis and have been conducted in a wide geographic distribution. The additional inclusion of 577.1 and 577.2 does not have the literature support to recommend use.
The authors declare no conflict of interest.
- The most common algorithm used was for acute pancreatitis and was based on the ICD-9-CM code 577.0. The PPV for this algorithm ranged from 40% to 97%, with the majority of data indicating a PPV of approximately 70%.
- The major limitations of the studies were sample heterogeneity; several studies were limited to high-risk populations (e.g., heavy alcohol use, dialysis, HIV, gallstones), whereas others studied typical risk populations, pediatric patients, or samples in the typical age range for this condition.
- Some studies included either ICD-9-CM code 577.0 acute pancreatitis or ICD-9-CM code 577.1 chronic pancreatitis, which may only minimally increase sensitivity of the algorithm and at the risk of reduced specificity.
- Use of ICD-9 code 577.1 may have particularly poor PPV as it can include both recurrent acute pancreatitis and chronic pancreatitis. Current definitions of recurrent acute pancreatitis, chronic pancreatitis, as well as the absence of an alternative diagnosis for abdominal pain and mild pancreatic enzyme elevations are problematic for use of administrative data for this outcome.
This work was supported by a contract from the US FDA through the Department of Health and Human Services Contract Number HHSF2232009100061.