Development and validation of a case definition to estimate the prevalence and incidence of cirrhosis in pan-Canadian primary care databases

Abstract

Aims:

To develop and validate case definitions to identify patients with cirrhosis and alcohol-related cirrhosis using primary care electronic medical records (EMRs) and to estimate cirrhosis prevalence and incidence in pan-Canadian primary care databases, between 2011 and 2019.

Methods:

A total of 689,301 adult patients were included with ≥1 visit to a primary care provider within the Canadian Primary Care Sentinel Study Network between January 1, 2017, and December 31, 2018. A subsample of 17,440 patients was used to validate the case definitions. Sensitivity, specificity, predictive values were calculated with their 95% CIs and then determined the population-level prevalence and incidence trends with the most accurate case definition.

Results:

The most accurate case definition included: ≥1 health condition, billing, or encounter diagnosis for International Classification of Diseases, Ninth Revision codes 571.2, 571.5, 789.59, or 571. Sensitivity (84.6; 95% CI 83.1%–86.%), specificity (99.3; 95% CI 99.1%–99.4%), positive predictive values (94.8; 95% CI 93.9%–95.7%), and negative predictive values (97.5; 95% CI 97.3%–97.7%). Application of this definition to the overall population resulted in a crude prevalence estimate of (0.46%; 95% CI 0.45%–0.48%). Annual incidence of patients with a clinical diagnosis of cirrhosis nearly doubled between 2011 (0.05%; 95% CI 0.04%–0.06%) and 2019 to (0.09%; 95% CI 0.08%–0.09%).

Conclusions:

The EMR-based case definition accurately captured patients diagnosed with cirrhosis in primary care. Future work to characterize patients with cirrhosis and their primary care experiences can support improvements in identification and management in primary care settings.

Lay summary

Cirrhosis is a final stage of any chronic liver disease and is associated with significant morbidity, health care utilization, and mortality. Worldwide cirrhosis is responsible for approximately 2 million deaths each year. Despite advancements in medicine, cirrhosis prevalence is rising, especially in younger age groups. This is likely related to increasing use of harmful alcohol consumption and obesity. This anticipated rise in disease burden of cirrhosis requires the development of an appropriate population-based chronic disease management and prevention plan. However, prior to that we need to determine an accurate estimation of the disease burden.

Primary care electronic medical records have been used extensively in disease surveillance, clinical outcome measurement, and quality of care improvement research. These have also shown to be useful in describing the key risk factors for cirrhosis, specifically BMI and alcohol use. The purpose of this study was to develop and validate a case definition that accurately identifies adult patients with cirrhosis and alcohol-related cirrhosis in primary care settings and estimate prevalence and incidence of cirrhosis in a primary care setting in Canada. This definition will be useful for future studies aimed at measuring the impact of care and burden of illness related to cirrhosis. We encourage methodological contributions aimed at improving the design and analysis of epidemiologic studies that impact population health.

Introduction

Cirrhosis is the common final stage of any chronic liver disease and is associated with significant morbidity, health care utilization, and mortality (1–3). Worldwide, more than 2 million deaths are attributed each year to liver cirrhosis and its complications including portal hypertension, liver failure, and hepatocellular carcinoma (2,4,5). It is widely accepted that, due to incomplete ascertainment of cirrhosis-related death the mortality is underestimated, and the true disease burden is significantly higher than reported (6).

Since the advent of highly effective medications able to cure chronic hepatitis C in the last decade, alcohol-related liver disease (ALD) and nonalcohol fatty liver diseases (NAFLD/NASH) have established themselves as the most common causes of cirrhosis in Europe and North America (7). Their prevalence is rising, especially in younger age groups, in parallel with a rising prevalence of harmful alcohol consumption and obesity (8,9). During the COVID-19 pandemic, sales of alcohol increased by 150% in some Canadian provinces and survey data indicated a 10%–25% increase in rates of alcohol consumption (10,11). Mortality for alcohol-related cirrhosis is high and expected to increase by 75% by 2040 if the current trends of high-risk drinking continues (12). The prevalence of obesity in adults has also risen dramatically in Canada, increasing three-fold over the last several years (13). Between 1985 and 2016, the prevalence of class II (BMI 35–39) and III (BMI >40) obesity increased by 455% (13). It is estimated that 1.9 million adult Canadians are class II or III obese (14) and at risk of developing NASH related liver cirrhosis. Even if only a relatively small proportion of individuals with high risk drinking patterns or obesity develop cirrhosis and its complications, the sheer number of the population at risk is anticipated to translate into a substantially higher prevalence of cirrhosis and cirrhosis-related health service utilization over the next decade and beyond (15,16). The anticipated raising disease burden of cirrhosis requires the development of an appropriate population-based chronic disease management and prevention plan, the first step of which is an accurate estimation of the time trends of incidence and prevalence of cirrhosis.

Electronic medical records (EMRs) are increasingly being used in primary care settings and now include large volumes of clinical data that are potentially useful for research. Primary care EMR data have been used extensively in disease surveillance, clinical outcome measurement, and quality of care improvement research (17). EMR data have also shown to be useful in describing the key risk factors for cirrhosis, specifically BMI and alcohol use (18). While selection bias is inherent to any population seen and captured in the database of specialized referral centres, the primary care EMR based data from a large national network such as that of the Canadian Primary Care Sentinel Study Network (CPCSSN) can more closely reflect the entire population (19,20). Using EMR data to correctly identify patients with cirrhosis (or any other diagnostic entity) requires the development of a case definition and the validation of case-finding diagnostic algorithms. Previous validation studies for cirrhosis are limited by small cohorts and patient population restricted to liver clinics, which has the potential to limit generalizability of the results (21–27). Studies utilizing primary care EMR data to determine prevalence and epidemiologic characteristics of cirrhosis in adults are also lacking.

The purpose of this study was to develop and validate a case definition that accurately identifies adult patients with cirrhosis and alcohol-related cirrhosis in primary care settings and estimate prevalence and incidence of cirrhosis in a primary care-based cohort.

Methods

Data sources

The CPCSSN is Canada’s first multi-disease surveillance system; it encompasses de-identified primary care EMR data from 13 practice-based research networks in eight Canadian provinces (19,28,29). These networks are located in Alberta, British Columbia, Manitoba, Newfoundland and Labrador, Nova Scotia, Ontario, Quebec, and the Northwest Territories. Fifteen hundred primary care providers have consented to participate in CPCSSN, representing ∼1.8 million patients. Data are extracted from the EMR of consenting providers semi-annually, de-identified, and aggregated into a single Canadian’s largest data repository to be used for research and surveillance (19). The extracted data include encounters, disease/health conditions, demographics, comorbidities, billing data, prescribed medications, laboratory results, selected risk factors, referrals, and procedures for primary care patients (28). International Classification of Diseases, Ninth Revision (ICD-9) is the current coding standard for diagnoses submitted for the purpose of remuneration in primary care across many EMRs in Canada.

Several case definitions have been developed and validated for various chronic conditions using the CPCSSN, including diabetes, chronic obstructive pulmonary disease, chronic kidney disease, and dementia (30–33). Provinces representation to the CPCSSN sample includes; Ontario (48.2%) followed by Alberta, Nova Scotia, Newfoundland, and Manitoba (15.8%, 14.1%, 9.5%, 8.2%, respectively), whereas Networks in British Columbia and Quebec contributed under 5% of the total sample (20). That said, people included in CPCSSN data are somewhat representative to the general population, although with higher representation of older individuals and women, and fewer young to middle aged men (20). This pattern is typical for most primary care study cohorts and may be explained by the fact that older people generally have greater health care needs than younger individuals and women tend to seek medical care more frequently than men (34–36).

Study design and population

This retrospective cross-sectional study used EMR data from the CPCSSN repository extracted in the first half of 2020. The study population consisted of de-identified records of all active adult patients. An “active” patient was defined as individual aged 18 years and over on January 1, 2017, with at least one encounter with a provider participating in the CPCSSN between January 1, 2017, and December 31, 2018. This method of using the 2-year contact period has shown to be effective in capturing the practice population (37). To establish the reference standard for cirrhosis, an experienced family physician clinician-scientist (AS) and a hepatologist (NF) with more than 10 years of experience on managing cirrhosis developed a list of key words, medications, and billing codes to identify patients with possible cirrhosis. These patients were reviewed by the hepatologist (NF) to assess presence or absence of a cirrhosis diagnosis based on available documentation in the record. We used a standardized data extraction form to collect details including diagnosis, diagnostic codes, medication data, and features of cirrhosis from the patients’ chart that supported the diagnosis of cirrhosis. This method of using the CPCSSN-processed data to identify the reference cases has been proven to be an acceptable substitute for a full chart review (38).

We constructed a validation cohort of 17,440 patients from the initial study population (Figure 1). The cohort comprised of cirrhosis cases (as identified by the reference standard identification approach as defined above) and cirrhosis non-cases. Cirrhosis non-cases were derived from a random sample of patient records without cirrhosis key words/codes in the CPCSSN EMR data. Medical students and family medicine residents assisted with the medical record review and confirmed the inclusion of non-cases. Medical students and family medicine residents attended a 1-hour training session specific to creation of reference set, additional guidance, and instruction was provided as needed by AS and LK.

Figure 1:

Selection of study population from the CPCSSN data repository

CPCSSN = Canadian Primary Care Sentinel Surveillance Network; EMR = Electronic medical record

Cirrhosis case definitions

An iterative process was used in the creation and testing of eight possible cirrhosis case definitions. Case definitions 1–5 were developed by the study team including a family physician (AS), hepatologist (NF), and researcher (LK) familiar with the CPCSSN data standards. These definitions used a combination of ICD-9 codes (used by primary care physicians for remuneration purposes in Canada), and textual data drawn from a number of sections within the EMR, including the problem and encounter diagnoses, billing, laboratory test results, and prescribed medications. Definitions were designed to range from being highly specific and less sensitive, to less specific and more sensitive (Table 1).

Table 1:

Estimates of validity for cirrhosis case definitions in the Canadian Primary Care Sentinel Surveillance Network (n = 17,440)

Case definition	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	ACC (95% CI)
Case definition 1 ≥1 ICD-9 codes 571.2, 571.5 in Health Condition table, billing, or encounter diagnosis	54.5 (52.5–56.5)	99.7 (99.5–99.7)	96.2 (95.1–97.1)	93.0 (92.8–93.3)	93.3 (92.9–93.7)
Case definition 2 ≥1 ICD-9 codes 571.2, 571.5, 789.59 in health condition table, billing, or encounter diagnosis	54.5 (52.6–56.5)	99.6 (99.5–99.7)	96.0 (94.8–96.9)	93.0 (92.8–93.3)	93.3 (92.9–93.7)
Case definition 3 1 ICD-9 code 571.2, 571.5, or 789.59 in health condition table OR ≥2 ICD-9 codes 571.2, 571.5, or 789.59 billing, or encounter diagnosis within 1 year	51.2 (49.2-53.2)	99.8 (99.7-99.8)	97.4 (96.3-98.1)	92.6 (92.3-92.9)	92.9 (92.5-93.3)
Case definition 4 ≥1 ICD-9 codes 571* excluding 571.0 and 571.6 in health condition table, billing, or encounter diagnosis	82.7 (81.1–84.2)	98.5 (98.3–98.7)	90.2 (89.0–91.3)	97.2 (97.0–97.4)	96.3 (96.0–96.6)
Case definition 5 ≥1 ICD-9 codes 571.2, 571.5, 789.59 in health condition table, billing, or encounter diagnosis OR ≥2 ICD-9 codes 155, 155.2, 197.7, 456, 456.1, 456.2, 456.21, 456.8, 789.2 in health condition table, billing, or encounter diagnosis	63.0 (61.0–64.9)	98.3 (98.1–98.5)	86.0 (84.4–87.5)	94.2 (93.9–94.5)	93.4 (93.0–93.7)
Case definition 6 ≥1 ICD-9 codes 571* or 571.6 excluding 571.0 in health condition table, billing, or encounter diagnosis OR ≥1 ICD-9 codes 789.2, 789.59, 155, 155.2, 197.7 in health condition table, billing or encounter diagnosis OR ≥1 ICD-9 codes 456* excluding 456.6, 456.5, or 456.4 in health condition table, billing, or encounter diagnosis OR ≥1 medication prescribed for ATC codes: A07EA03, A07EA06, L04AX01	84.0 (82.5–85.5)	96.9 (96.6–97.2)	81.6 (80.2–82.9)	97.4 (97.1–97.6)	95.1 (94.8–95.4)
Case definition 7 ≥1 ICD-9 codes 571* excluding 571.0 and 571.6 in health condition table, billing, or encounter diagnosis OR ≥1 ICD-9 codes 789.59 in health condition table, billing, or encounter diagnosis	82.9 (81.3–84.3)	98.5 (98.3–98.7)	90.1 (88.8–91.2)	97.2 (97.0–97.5)	96.3 (96.0–96.6)
Case definition 8 ≥1 ICD-9 codes 571.2, 571.5, 789.59, or 571 (no qualifier) in health condition table, billing, or encounter diagnosis	84.6 (83.1–86.0)	99.3 (99.1–99.4)	94.8 (93.9–95.7)	97.5 (97.3–97.7)	97.2 (96.9–97.4)

Definitions included ICD-9 codes: 571 (chronic liver disease and cirrhosis), 571.2 (alcoholic cirrhosis of liver), 571.5 (cirrhosis of liver without mention of alcohol), 789.59 (other ascites), 155 (malignant neoplasm of liver and intrahepatic bile ducts), 197.7 (liver, specified as secondary), 456 (varicose veins of other sites), and 789.2 (splenomegaly). Case definition 6 also included medications using ATC codes: A07EA03 (prednisone), A07EA06 (budesonide), and L04AX01 (azathioprine).

^* Indicates all subcodes are included.

ATC = Anatomical Therapeutic Chemical; ICD = International Classification of Diseases; NPV = Negative predictive value; PPV = Positive predictive value

Covariates

We used specific ICD-9 codes to identify individuals for having cardiovascular disease. The risk factor table in the CPCSSN captured current alcohol use and smoking (18,39). Patients were identified as having alcohol use disorder/dependence with any occurrence of related ICD-9 codes in the billing, health condition table or encounter diagnosis fields (details of ICD-9-CM codes, supplemental S1). Demographic information (age and sex) was obtained from the patient demographics table. Urban clinic location was determined using the clinic’s three-digit postal code. Annual visit frequency and annual medication were determined by calculating the mean from the total for each year (2017, 2018, and 2019).

Statistical analysis

The study population was described using frequencies and percentages. Validity of the cirrhosis case definitions were assessed using estimates of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy. Accuracy was computed by dividing the sum of the true positive and true negative cases by the sum of true positive, true negative, false negative, and false positive cases. The 95% CIs were computed and are based on the asymptotic standard error and a critical value from the standard normal distribution.

We applied the optimal case definition to the entire population of CPCSSN and observed the lifetime prevalence and created annual cohorts of active patients to assess the trends of incidence from 2011 to 2019 for those aged 18 or above. This yearly count was used to determine the observed incidence, defined as the proportion of the CPCSSN adult population with newly diagnosed cirrhosis (first met the cirrhosis criteria) each year stratified by sex. Estimates of 95% CIs were computed. Statistical analyses were conducted using SAS V9.4 (SAS Institute Inc, Cary, NC, USA). This study was approved by the Health Research Ethics Board at the University of Manitoba (H2017:257 (HS1053)) and the CPCSSN Standing Committee on Research and Surveillance. Informed consent to extract EMR data was provided by primary care providers. The CPCSSN data are de-identified at the source and therefore individual patient consent cannot be obtained.

Results

Figure 1 depicts the initial study population of 1,493,516 individuals. Among them 683,325 fulfilled the inclusion criteria of age (18 years or older) and had at least one recorded visit between January 1, 2017, and December 31, 2018. We arrived at the final validation cohort of 17,440 individuals. Data are reported from regional networks in British Columbia, Alberta, Manitoba, Ontario, Quebec, Nova Scotia, and Newfoundland.

The performance of the case definitions for the diagnosis for cirrhosis over a 2-year period is shown in Table 1. Case definition 8 had the best performance; sensitivity (84.6; 95% CI 83.1–86.0) specificity (99.3; 95% CI 99.1–99.4), PPV (94.8; 95% CI 93.9–95.7), and NPV (97.5; 95% CI 97.3%–97.7%). It included any occurrence of ICD 9 codes: 571.2, 571.5, 789.59, or 571 in the billing, health condition table, or encounter diagnosis table in CPCSSN data. The highly specific case definition 1 had low sensitivity (54.5; 95% CI 52.5–56.5). Including all codes 571 (excluding 571.0 and 571.6) in case definition 4 greatly improved sensitivity (82.7; 95% CI 81.1–84.2), however the PPV dropped to 90.2 (95% CI 89.0–91.3).

Alcohol-related cirrhosis

We identified 686 patients with alcohol-related cirrhosis based on alcohol use disorder/dependence diagnosis codes and absence of other causes of cirrhosis diagnosis codes, medications, and key words. We selected 4,185 with no alcohol use or cirrhosis diagnosis. This created a validation cohort of 4,871 patients that was used to test four case definitions for alcohol-related cirrhosis (Supplemental S2). Use of a specific ICD-9 code for alcohol-related cirrhosis of the liver (571.2) produced a sensitivity estimate of 84.2 (95% CI 80.0–87.9), specificity estimate of 91.4 (95% CI 90.6–92.2), PPV estimate of 43.6; (95% CI 41.0–46.2), NPV estimate of 98.7 (95% CI 98.3–98.9), and accuracy estimate of 90.9 95% CI 90.1–91.7). This definition performed better than other case definitions that included patients with cirrhosis (case definition 8) and alcohol dependence/abuse ICD-9 codes (291, 303, 305.0) (Supplemental S1).

Estimates of cirrhosis prevalence and incidence

Using the best performing case definition 8 and extrapolating to the 2-year active patient population captured within CPCSSN results in an estimated lifetime prevalence estimate of 0.46% (95% CI 0.45–0.48) in the primary care population. Figure 2 provides annual incidence estimates by sex. Although male patients had a consistently higher incidence estimate of cirrhosis, the incidence rate among female patients increased significantly over the study period. Among men incidence rose from 0.08%, (95% CI 0.07%–0.1%) to 0.11% (95% CI 0.1%–0.13%) between 2011 and 2019, whereas incidence among female rose from 0.03% (95% CI 0.02%–0.04%) to 0.07% (95% CI 0.06%–0.07%) (p = 0.02). Using annual incidence rates, we assessed the cumulative burden of cirrhosis which increased from 0.09% (95% CI 0.08%–0.1%, 466/504492 active patients between 2006 and 2011) in 2011 to 0.27% (95% CI 0.26%–0.28%, 3267/1213131 active patients between 2014 and 2019) in 2019. On average patients were 59.2 (SD 12.8) years of age when cirrhosis was first documented in EMR data.

Figure 2:

Annual Incidence of cirrhosis by sex (2011–2019) among patients with at least one visit to a primary care provider participating in the Canadian Primary Care Sentinel Surveillance Network between January 1, 2017, and December 31, 2019

Characteristics of cirrhosis cases and non-cirrhosis cases among the 2-year active patient population are shown in Table 2. The 3,201 patients captured with the best performing liver cirrhosis case definition 8 were significantly more likely to be male (56.4% versus 43.7%, p < 0.0001) and had an older mean age (62.3 [SD 13.3] versus 52.2 [SD 19.4] p < 0.0001) compared to patients without cirrhosis. Patients with cirrhosis were significantly more likely to have documentation of smoking, alcohol, or drug use in the EMR (ie, smoking 11.8% versus 4.2%, alcohol use 24.4% versus 2.4%, drug use 13.0% versus 3.5%). Diabetes, chronic kidney disease, and cardiovascular disease were all significantly more common in the population with cirrhosis (33.7% versus 13.2%, p < 0.0001, 13.5% versus 5.6%, p < 0.0001 and 21.9% versus 11.1%, p < 0.0001 respectively). In addition, 26.6% of patients (852) had a physical condition related to alcohol use, compared to 0.5% (3,205) in patients without cirrhosis.

Table 2:

Characteristics of patients with at least one visit to a primary care provider participating in the Canadian Primary Care Sentinel Surveillance Network between January 1, 2017, and December 31, 2019

Patient characteristic	Without cirrhosis, no. (%)*; n = 686,100	With cirrhosis, no. (%)*; n =3,201	p-value
Patient sex (male)	299,499 (43.7)	1,802 (56.4)	<0.0001
Patient age, mean (SD)	52.2 (19.4)	62.3 (13.3)	<0.0001
Urban (%) versus rural residency	515,138 (80.4)	2,421 (80.8)	0.6342
Annual physician visits, mean (SD)	2.8 (3.4)	5.9 (6.0)	<0.0001
Chronic kidney disease	38,250 (5.6)	432 (13.5)	<0.0001
Diabetes	90,604 (13.2)	1,080 (33.7)	<0.0001
Cardiovascular disease	76,363 (11.1)	701 (21.9)	<0.0001
Diagnosed smoking	28,817 (4.2)	376 (11.8)	<0.0001
Diagnosed alcohol	16,103 (2.4)	781 (24.4)	<0.0001
Physical condition related to alcohol use	3,205 (0.5)	852 (26.6)	<0.0001
Drug use	24,169 (3.5)	415 (13.0)	<0.0001

ICD = International Classification of Diseases

Among those with cirrhosis, 686 patients (21.4%) met the definition of alcohol-related cirrhosis. The annual incidence of alcohol-related cirrhosis was stable over the study period (48 patients [0.01%] in 2011 to 141 patients [0.02%] in 2019).

Discussion

This study developed and validated a method for identifying individuals with known cirrhosis and alcohol-related cirrhosis using a pan-Canadian reference standard from primary care EMRs in seven Canadian provinces. We believe this is the first primary care EMR-based validation study for cirrhosis identification. The case finding algorithms demonstrated excellent specificity, NPV, PPV, and accuracy as well as good sensitivity in the CPCSSN databases for identifying patients with and without cirrhosis. Depending on the goals of future study, these algorithms can be selected based on high sensitivity to identify maximum probable cirrhosis cases and high specificity algorithm to avoid detecting false positive disease cases.

Case definitions have previously been validated for various chronic conditions in CPCSSN database, but not for cirrhosis or chronic liver disease. Our study contributes to the literature for demonstrating the utility of CPCSSN case defining algorithms to accurately identify patients for research and clinical purposes within Canadian primary care settings. Thus far only one Canadian study from Ontario has validated a diagnosis of cirrhosis using health administrative data (26). This validation study used data from two hepatology clinics and observed that a single hospital diagnosis code for cirrhosis, including 572.2, was specific for cirrhosis (91%–96%) depending on sub-cohort, but not as sensitive (57%–77%) (26). Further, this study demonstrated that a single physician visit code for cirrhosis was sensitive (98%–99%), but not as specific (66%–78%) (26). While a limited diagnosis code may be adequate for specialized clinics, more broad consideration may be needed for primary care settings as we have shown.

CPCSSN definition for cirrhosis (PPV 98%) was comparable to, or better than, other definitions developed in similar databases in Europe and North America. Some definitions (40) used cirrhosis diagnosis codes only, whereas others (41,42) also included varices codes. For example, Ratib et al (42) included procedure codes for treatment of varices in addition to cirrhosis codes (ICD-10 codes) in United Kingdom’s Clinical Practice Research Datalink. The PPV estimates were 90%; however, sensitivity, specificity, and NPV were not reported. Similarly, a diagnostic algorithm for cirrhosis in the Danish registry included one or more inpatient or outpatient ICD-10 diagnostic codes of cirrhosis or cirrhosis-related conditions with PPV of 71%, other validity metrics were not reported (43). A study sample of electronic health record data from Veteran Affairs used algorithm that included one inpatient or ≥2 outpatient codes for ascites, peritonitis, or variceal hemorrhage with poor sensitivity (20%) but high PPV estimates (91%) (22).

Our EMR case definitions for alcohol-related cirrhosis had sensitivity and specificity above 85%, however all case definitions also had high number of false positives, resulting in low PPV. In general, an algorithm with high specificity is preferable to identify a cohort of patients as it maximizes the likelihood that individuals have the condition of interest. Definition 1 (specific ICD-9 code 571.2) provides a balance between sensitivity and PPV and could serve as an important tool to identify alcohol-related cirrhosis cases for disease surveillance in Canadian primary care practices.

Application of the cirrhosis case definition to national CPCSSN data provided a crude prevalence of 0.5% among adult population who were seen by a primary care provider in Canada. Between 2011 and 2019, the incidence of patients with a clinical diagnosis of cirrhosis rose from 0.05 to 0.09 with larger increases in annual incidence among females than in men. This observation corresponds to the recent study that observed that the incidence of cirrhosis among women is increasing faster possibly related to increase in alcohol related disease in women (44). Prevalence and incidence estimates for cirrhosis may vary by population characteristics (eg, age and sex distribution) and burden of risk factors (eg, high risk alcohol use and metabolic syndrome) (13,45–47). However, methodological difference can also impact on these estimates such as difference on the data sources, diagnosis codes, and study population and case definition (48). For example, Ontario reported a higher prevalence of cirrhosis of 0.84% in 2016 using health administrative data (7), whereas Tonelli et al reported cirrhosis prevalence at 0.2% in Edmonton in 2008–2009 using hospitalizations claims and ambulatory care data and 0.1% when only hospitalization claims data were used. Our EMR based definitions may underestimate the true prevalence of cirrhosis and alcohol related cirrhosis in patients cared for in primary care settings as cirrhotic patients may be referred to specialist care for the management of cirrhosis.

The validity of the case definition is further supported by the characteristics of the CPCSSN cirrhosis population. As expected, patients with cirrhosis were significantly more likely to be diagnosed with diabetes, chronic kidney, and cardiovascular diseases, and substance use abuse/dependence, consistent with findings from population-based data sources (49). The prevalence of diabetes, chronic kidney, and cardiovascular diseases were noted to be within the bounds of prior estimates of the prevalence of these conditions (49). This may suggest that population with diagnosed cirrhosis is being managed in primary care settings for related comorbidities, but it is unclear if they were receiving active cirrhosis management by their primary care provider. Although cirrhosis comorbidities are often related to its underlying etiology such as metabolic syndrome and alcohol use disorder (50), primary care providers report challenges to offer cirrhosis management and specialty care coordination (51). We found that one-third of patients with cirrhosis had an alcohol-related diagnosis in EMR data compared to only 2.5% of patients without a cirrhosis diagnosis. Other literature has shown that comorbidity affects the prognosis of cirrhosis patients (49). That said, the role of primary care providers in the management of cirrhosis patients is vital for improving overall quality and long-term survival.

Our study has several limitations. First, cirrhosis is an insidious disease and majority of the patients remain asymptomatic and have normal liver enzymes (52). This poses a particular challenge for identifying patients with mild or less severe liver disease using EMR data. Second, primary care providers largely use the EMR for clinical documentation and may be less focused on secondary use of the data for health research or disease surveillance (53). Therefore, the accuracy of specific ICD-9 codes has possible impacts on algorithm performance and prevalence estimates. Third, the use of short-diagnostic text and free-text in the EMR contain spelling errors and medical jargon that may not necessarily be readily captured by simple cleaning algorithms that were applied (28). Fourth, our case definition for alcohol-related cirrhosis resulted in low PPV due to a large number of false positives. Although some of them may be true positive and alcohol-related cirrhosis was not documented in the short-diagnostic text in the EMR. It is also possible that not all patients with cirrhosis sought care from their primary care provider for the condition and some of the documentation could be found in other sources (eg, hospital data). Lastly, issues with estimating incidence, as EMR record does not contain explicit notations that can distinguish the first onset of disease from the follow-up visits for continuing care or simply the addition of a new patient to the practice that has the condition. Despite these limitations, there are no other practical methods available to achieve this large scale, generalizable, and efficient research from primary are settings besides using routine health care records. Future studies could employ data linkage with specialists (ie, hepatology) or hospital-based data to increase the sensitivity and specificity of prevalence and incidence estimates.

Conclusion

This study is the first to use primary care data to validate case definitions for cirrhosis in Canada. Our cirrhosis definition was satisfactorily specific and sensitive for use to expand the understanding of the prevalence and incidence of cirrhosis patients managed within primary care settings. This definition will be useful for future studies aimed at measuring the impact of care and burden of illness related to cirrhosis.

Funding Statement

This study is supported by Advanced Analytics Grant from IBM and Canadian Institute for Military and Veteran Health Research.

Contributions:

Conceptualization, L Kosowan, A Singer, N Faisal; Data curation, L Kosowan, A Singer, H Zafari, F Zulkernine; Methodology, L Kosowan, A Singer,; Writing – Original Draft, N Faisal; Writing – Review & Editing, L Lix, A Mahar, H Singh, E Renner, and A Singer.

Ethics Approval:

This study was approved by Health Research Ethics Board at the University of Manitoba (H2017:257 (HS1053)) and the CPCSSN Standing Committee on Research and Surveillance.

Informed Consent:

N/A

Funding:

This study is supported by Advanced Analytics Grant from IBM and Canadian Institute for Military and Veteran Health Research.

Disclosures:

The opinions expressed in this paper are those of the authors and do not necessarily represent those of the funding agency. The study sponsor had no role in study design, the collection, analysis, and interpretation of data, the writing of the report; and the decision to submit the manuscript for publication. H Singh has been on advisory boards or consulted to Amgen Canada, Roche Canada, Sandoz Canada, Takeda Canada, Pendopharm Inc and Guardant Health, Inc. N Faisal received Mitacs Accelerate funds internship.

Peer Review:

This manuscript has been peer reviewed.

Animal Studies:

N/A