Abstract
Background:
Ascites is associated with significantly increased morbidity, mortality, and healthcare costs. Large population studies are necessary to determine the burden and impact of ascites, however ascites ICD-10 codes perform poorly in the identification of patients.
Methods:
We utilized three independent retrospective cohorts at the University of Michigan (cohorts 1 and 2) and Duke University (cohort 3). Cohort 1: Child A5-6 patients followed up to 10 years (n=150); cohort 2: Child A5-B7 patients with portal hypertension followed for up to 1 year (n=65); cohort 3: cross-sectional cohort of patients evaluated for liver transplant (n=100). We computed performance characteristics for ascites-related ICD-10 codes (K70.31, K70.11, K71.51, R18.8) as well as loop and/or potassium sparing diuretics.
Results:
315 patients were included across three cohorts. Algorithms including any ascites code provided better sensitivity and equivalent specificity to R18.8 alone for all cohorts. In cohort 2, we found that loop diuretics, potassium-sparing diuretics, and a combination of both with a cirrhosis code were highly sensitive (82.3% for each) and specific (89.1%−93.5%). In contrast, ascites codes were insensitive. In patients with moderate-severe ascites, a combination of recorded diuretics showed high sensitivity and specificity (95.2% and 86.8%). In Cohort 3’s transplant evaluation patients, we found that loop diuretics, potassium-sparing diuretics, and a combination of both with a cirrhosis code were highly sensitive (90.4%, 78.8% and 75.0% respectively) and specific (85.0%, 90.0% and 95.0% respectively). For moderate-severe cirrhosis, loop diuretics and R18.8 showed higher sensitivity (77.8%) and specificity (88.9%) respectively.
Conclusion:
Diuretic records improve the accuracy of ascites ICD-10 codes the identification of ascites. This method for identifying ascites should be used in future large dataset studies.
Keywords: Liver Disease, Ascites, Cirrhosis, Administrative Codes, Diuretics
Introduction
Ascites is defined by the appearance of fluid in the peritoneal cavity and is a common complication of cirrhosis, occurring in as many as 60% of patients within 10 years.3 It results in a substantial symptom burden and increased healthcare utilization.4 Once patients develop ascites, mortality ranges from 15–20% in 1 year to up to 80% at 5 years.5 For these reasons a correct estimation of the disease burden is necessary to assess the personal and public health footprint of ascites, something which can only be achieved using large, administrative datasets.6
Administrative databases include population information that are routinely collected during hospital, clinic, laboratory, or pharmacy visits across large geographic areas.7,8 However, administrative data are only useful when the methods applied to identify patients are accurate and externally valid.7 Validation is typically achieved by comparing candidate diagnostic codes against an accepted clinical reference standard, such as physician chart review.9,10 Previous studies regarding cirrhosis and its complications have relied on International Classification of Diseases (ICD)-9 diagnosis codes.11–14 However, ICD-10 codes for cirrhosis and cirrhosis-related complications have demonstrated variable sensitivity and specificity.7 It is unknown whether the ICD-10 codes for ascites are accurate. We recently showed that using medications specific for the cirrhotic complication, i.e. lactulose for encephalopathy, can improve case identification. Such data regarding ascites are lacking.
Herein, we evaluate the performance of diagnostic algorithms for the detection of patients with ascites using combinations of ICD-10 codes and diuretic prescriptions in 3 cohorts of patients with cirrhosis at two US medical centers.
Methods
Patient Selection
We conducted our study using three independent cohorts of patients at the University of Michigan and Duke University. Cohort 1, a retrospective cohort of 150 patients with compensated patients followed for up to 10 years from the University of Michigan Hepatology Clinic, (years). Cohort 2, a prospective cohort with 65 patients with Child A5-B7 cirrhosis and portal hypertension followed for 1 year from the University of Michigan Hepatology Clinic, (year). And cohort 3, a retrospective cohort of 100 patients randomly selected from the list of patients evaluated for liver transplant from Duke University Liver Transplant Clinic, (2018–2020). Patients were included if aged ≥18 years with a confirmed diagnosis of cirrhosis based on imaging or histological criteria and documented as such by a board certified hepatologist. We excluded patients with prior liver transplant and metastatic cancer.
Study Data Collection
We performed a manual chart review gathering demographic characteristics (age, sex, race), etiology of cirrhosis, MELD-Na score, Child-Pugh score, documented history of ascites and/or hydrothorax, paracentesis or thoracentesis in last 3 months and history of transjugular intrahepatic portosystemic stent shunt (TIPS). Ascites was graded as follows: 1) none, 2) minimal (small volume on imaging or resolved with diuretics within 3 months), 3) moderate (seen on examination, moderate volume on imaging), and 4) severe (symptomatic/tense and/or requiring paracentesis).
For the first two cohorts at the University of Michigan Hepatology Clinic, we surveyed all cirrhosis and ascites ICD-10 billing-codes generated for each patient up to last visit. For the cohort study at Duke University Liver Transplant Clinic, we surveyed all cirrhosis and ascites ICD-10 billing-codes generated from the clinic note before decision was made regarding proceeding or not with liver transplantation. Cirrhosis ICD-10 codes included: K74.6 (other and unspecified cirrhosis of liver), K70.3 (alcoholic cirrhosis of liver), K74.4 (secondary biliary cirrhosis) and K74.5 (biliary cirrhosis, unspecified). Ascites ICD-10 codes included: K70.31 (alcoholic cirrhosis of liver with ascites), K70.11 (alcoholic hepatitis with ascites), K71.51 (toxic liver disease with chronic active hepatitis with ascites) and R18.8 (other ascites). For all three cohorts, we evaluated medication records for prescription and doses of loop diuretics (furosemide, torsemide and bumetanide) and potassium-sparing diuretics (spironolactone, eplerenone and amiloride).
Outcomes
We first determined the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for all ascites-related ICD-10 codes and recorded diuretics (loop + potassium-sparing diuretics) to identify grades 1–4 of ascites. We then defined the performance of diagnostic algorithms that included ascites-related ICD-10 codes or recorded diuretics with and without a cirrhosis code to identify patients with cirrhosis complicated by grades 1–4 of ascites. Lastly, we evaluated the performance of diagnostic algorithms that included a combination of ascites-related ICD-10 codes and recorded diuretics with and without cirrhosis code to identify grades 1–4 of ascites.
Data Analysis
Characteristics of patients in each validation cohort were described using univariate statistics. Different algorithms including ascites ICD-10 codes and/or recorded diuretics were evaluated for sensitivity, specificity, PPV and NPV. Performance measures were reported with their 95% confidence intervals. Sensitivity indicated the probability of ascites ICD-10 and/or recorded diuretics being present in those with ascites, whereas specificity indicated the probability of ascites ICD-10 codes and/or recorded diuretics would be absent in those without ascites. Sensitivity and specificity are not affected by disease prevalence. PPV indicated the probability that a patient with ascites ICD-10 and/or recorded diuretics had ascites, whereas NPV indicated the probability that a patient without a code and/or recorded diuretics did not have ascites. Data from manual chart review were the gold standard compared with codes from the databases. This study was exempted from Institutional Review Board review at Michigan (HUM00202050) and Duke Health (PRO00107357).
Results
Cohort Descriptions
Baseline characteristics for each cohort are presented in Table 1. Overall, 315 patients with similar age, sex, and race were included. Cohort 1, compensated patients followed retrospectively for up to 10 years, included 150 patients. The most common etiology of cirrhosis was alcohol (28%), followed closely by hepatitis C (24.7%) and non-alcoholic steatohepatitis (22.0%). MELD-Na and Child-Pugh scores were 12.8 +/− 6.20 and 6.29 +/− 1.66 respectively. Overall, 76.0% of patients had ascites during follow-up, 14.0% moderate-severe ascites; 60.0% of patients were prescribed a loop diuretic and 55.4% a potassium-sparing diuretic.
Table 1.
Characteristics of Patients in Each Validation Cohort
| Cohort 1 Retrospective compensated cirrhosis |
Cohort 2 Prospective Child A5-B7 cirrhosis |
Cohort 3 Retrospective liver transplant evaluation |
|
|---|---|---|---|
| (n=150) | N (%) or Median (IQR) (n=65) | (n=100) | |
| Age, years | 61.1 | 62.5 | 56.3 |
| Sex, male | 85 (56.7%) | 47 (72.3%) | 62 (62.0%) |
| Race, white/Caucasian | 135 (90.0%) | 57 (87.7%) | 89 (89.0%) |
| Etiology of cirrhosis | |||
| Alcohol | 42 (28.0%) | 7 (10.7%) | 11 (11.0%) |
| Hepatitis C | 37 (24.7%) | 30 (46.1%) | 12 (12.0%) |
| NASH | 33 (22.0%) | 13 (20.0%) | 29 (29.0%) |
| Autoimmune hepatitis | 7 (4.6%) | 6 (9.2%) | 5 (5.0%) |
| Cryptogenic | 7 (4.6%) | 2 (3.1%) | 3 (3.0%) |
| Primary biliary cholangitis | 10 (6.7%) | 3 (4.6%) | 4 (4.0%) |
| Primary sclerosing cholangitis | 3 (2.0%) | 0 (0.0%) | 11 (11.0%) |
| Other | 11 (7.4%) | 4 (6.1%) | 25 (25.0%) |
| MELD-Na score at clinic visit | 12.8 | 11.9 | 18.6 |
| Child-Pugh score at clinic visit | 6.3 | 6.1 | 8.7 |
| Documented ascites/hydrothorax | 114 (76.0%) | 21 (32.3%) | 64 (64.0%) |
| Recent paracentesis/thoracentesis | 18 (12.0%) | 7 (10.7%) | 27 (27.0%) |
| History of TIPS | 8 (5.4%) | 4 (6.1%) | 6 (6.0%) |
| Moderate-Severe Ascites | 21 (14%) | 10 (15.4%) | 41 (41%) |
| Ascites ICD-10 billing code (K70.11, K71.51, R18.8, K70.31) | 76 (50.7%) | 13 (20.0%) | 31 (31.0%) |
| Cirrhosis/Ascites ICD-10 billing codes | |||
| K74.6 - Other and unspecified cirrhosis of liver | 115 (76.7%) | 59 (90.8%) | 57 (57.0%) |
| K70.3 - Alcoholic cirrhosis of liver | 41 (27.3%) | 9 (13.8%) | 3 (3.0%) |
| K74.4 - secondary biliary cirrhosis | 0 (0.0%) | 1 (1.5%) | 0 (0.0%) |
| K74.5 - Biliary cirrhosis, unspecified | 2 (1.4%) | 1 (1.5%) | 0 (0.0%) |
| K70.11- Alcoholic hepatitis with ascites | 1 (0.6%) | 0 (0.0%) | 0 (0.0%) |
| K71.51- Toxic liver disease ascites | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) |
| R18.8 - Other ascites | 60 (40.0%) | 11 (16.9%) | 21 (21.0%) |
| K70.31 Alcoholic cirrhosis of liver with ascites | 25(16.6%) | 4 (6.1%) | 11 (11.0%) |
| Loop Diuretic | 90 (60.0%) | 20 (30.7%) | 57 (57.0%) |
| Loop Diuretic Type | |||
| Furosemide | 75 (83.4%) | 16 (80.0%) | 43 (43.0%) |
| Torsemide | 2 (2.2%) | 0 (0%) | 12 (12.0%) |
| Bumetanide | 13 (14.4%) | 4 (20%) | 2 (2.0%) |
| Potassium-sparing Diuretics | 83 (55.4%) | 17 (26.1%) | 51 (51.0%) |
| Potassium-sparing Diuretic type | |||
| Spironolactone | 76 (91.6%) | 15 (88.2%) | 47 (47.0%) |
| Eplerenone | 7 (10.4%) | 1 (5.8%) | 0 (0.0%) |
| Amiloride | 0 (0.0%) | 1 (5.8%) | 2 (2.0%) |
Cohort 2, Child A5-B7 patients followed prospectively for 1 year, included 65 patients. About half of the cohort had cirrhosis due to hepatitis C (46.1%). MELD-Na and Child-Pugh scores were 11.9 +/− 5.73 and 6.09 +/− 1.81 respectively. Overall, 32.3% had incident ascites, 15.4% moderate-severe; 30.7% of patients were prescribed a loop diuretic and 26.1% a potassium-sparing diuretic.
Cohort 3, evaluated cross-sectionally at the time of transplant evaluation, included 100 patients. The most common etiology of cirrhosis was non-alcoholic steatohepatitis (29.0%) followed by hepatitis C (12.0%) and alcohol (11.0%). MELD-Na and Child-Pugh scores were 18.65 +/− 6.95 and 8.68 +/− 2.2 respectively. Overall, 64.0% of patients had ascites, 41.0% moderate-severe; 57.0% of patients were prescribed a loop diuretic and 51.0% a potassium-sparing diuretic.
Performance of Identification Algorithms: Cohort 1 (Retrospective 10-year review)
Performance characteristics of ascites ICD-10 codes and recorded diuretics with and without a cirrhosis code are presented in Table 2–3, Supplementary Table 1. In cohort 1, we found that ascites codes and recorded diuretics had fair sensitivity (68.2% and 62.1%, respectively) and specificity (63.7% and 66.2, respectively). However, R18.8 and a diagnostic algorithm including a combination of ascites codes and recorded diuretics were insensitive. Meanwhile, PPV were high for all the algorithms (84.3%−89.2%), but NPV were extremely low for all of them (31.4%−44.6%) and there was no improvement with the inclusion of a cirrhosis code (Table 2 versus Supplementary Table 1). In patients with moderate-severe ascites,(Table 3) any ascites code showed highest sensitivity (90%) and R18.8 showed the highest specificity (67.5%). PPV (16.0%−29.3%) and NPV (89.2%−97.8%) for all algorithms.
Table 2:
Performance of ICD-10 Codes and Medications for the Identification of Persons with Any Ascites
| Codes/Medications | Cohort | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) |
|---|---|---|---|---|---|
| R18.8 | Cohort 1 Retrospective compensated cirrhosis |
57.6% (44.8% - 69.7%) | 75.0% (64.1% - 84.0%) | 87.5% (82.0% - 91.5%) | 36.0% (29.9% - 44.1%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
47.1% (22.9% - 72.2%) | 95.6% (85.2% - 99.5%) | 82.3% (52.2% - 95.2%) | 80.8% (72.8% - 86.9%) | |
| Cohort 3 Retrospective liver transplant evaluation |
40.4% (27.0% - 54.9%) | 100.0% (83.1%, 100%) | 100.0% | 39.2% (34.0% - 44.6%) | |
| Any ascites code (K70.3, R18.8, K70.11, K71.51) | Cohort 1 Retrospective compensated cirrhosis |
68.2% (55.5% - 79.1%) | 63.7% (52.2% −74.2%) | 85.1% (80.4% - 88.9%) | 39.6% (30.8 – 49.2%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
52.9% (27.8% - 77.0%) | 93.7% (82.8% - 98.7%) | 78.4% (52.6% - 92.2%) | 82.3% (73.6% - 88.5%) | |
| Cohort 3 Retrospective liver transplant evaluation |
58.5% (44.1% - 71.8%) | 100.0% (82.3%, 100%) | 100.0% | 46.3% (38.5% - 54.3%) | |
| Loop | Cohort 1 Retrospective compensated cirrhosis |
80.3% (68.7% - 89.1%) | 55.0% (43.5% - 66.1%) | 84.4% (80.6% - 87.7%) | 47.8% (35.1% - 60.7%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
82.3% (56.6% - 96.2%) | 89.1% (76.4% - 96.3%) | 76.4% (57.9% - 88.4%) | 92.2% (80.7% - 97.0%) | |
| Cohort 3 Retrospective liver transplant evaluation |
90.4% (78.9% - 96.8%) | 85.0% (62.1% - 96.8%) | 94.0% (84.6% - 97.8%) | 77.3% (59.1% - 88.9%) | |
| K-sparing | Cohort 1 Retrospective compensated cirrhosis |
71.2% (58.7% - 81.7%) | 57.5% (45.9% - 68.5%) | 83.6% (79.1% - 87.3%) | 39.6% (30.0% - 50.0%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
82.3% (56.5% - 96.2%) | 93.5% (82.1% - 98.6%) | 84.4% (63.9% - 94.3%) | 92.5% (81.5% - 97.2%) | |
| Cohort 3 Retrospective liver transplant evaluation |
78.8% (65.3% - 88.9%) | 90.0% (68.3% - 98.7%) | 95.3% (84.5% - 98.7%) | 62.1% (48.7% - 73.8%) | |
| Loop+ K-sparing | Cohort 1 Retrospective compensated cirrhosis |
62.1% (49.3% - 73.8%) (49.34% - 73.78%) |
66.2% (54.8% −76.4%) | 84.9% (79.6% - 88.9%) | 36.5% (28.9% - 44.8%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
82.3% (56.6% - 96.2%) | 93.0% (80.9% - 98.5%) | 83.5% (62.4% - 93.9%) | 92.5% (81.4% - 97.1%) | |
| Cohort 3 Retrospective liver transplant evaluation |
75.0% (61.0% - 85.9%) | 95.0% (75.1% - 99.9%) | 97.5% (85.1% - 99.6%) | 59.4% (47.4%, 70.3%) | |
| Loop + K-sparing + Code | Cohort 1 Retrospective compensated cirrhosis |
42.4% (30.3% - 55.2%) | 83.7% (73.8% - 91.0%) | 88.8% (81.8% - 93.4%) | 32.3% (27.5% - 37.5%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
52.9% (27.8% - 77.0%) | 93.5% (82.1% - 98.6%) | 77.7% (51.6% - 91.9%) | 82.2% (73.6% - 88.5%) | |
| Cohort 3 Retrospective liver transplant evaluation |
39.2% (25.8% - 53.9%) | 100.0% (83.9%, 100.0%) | 100.0% | 40.4% 35.2% - 45.8% |
NP = negative predictive value, PPV = positive predictive value.
Table 3:
Performance of ICD-10 Codes and Medications for the Identification of Persons with Moderate-severe Ascites
| Codes/Medications | Cohort | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) |
|---|---|---|---|---|---|
| R18.8 | Cohort 1 Retrospective compensated cirrhosis |
85.0% (62.1% - 96.8%) | 67.5% (58.5% - 75.5%) | 29.3% (23.3% - 36.2%) | 96.6% (90.8% - 98.8%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
60.0% (26.2% - 87.8%) | 96.2% (87.0% - 99.5%) | 77.7% (45.0% - 93.7%) | 91.6% (83.7% - 95.9%) | |
| Cohort 3 Retrospective liver transplant evaluation |
47.2% (30.4% - 64.5%) | 88.9% (73.9% - 96.9%) | 80.9% (61.3% - 91.9%) | 52.9% (44.7% - 60.9%) | |
| Any ascites code (K70.3, R18.8, K70.11, K71.51) | Cohort 1 Retrospective compensated cirrhosis |
90.0% (68.3% - 98.7%) | 55.6% (46.4% - 64.4%) | 24.3% (20.1% - 29.1%) | 97.2% (90.3% - 99.2%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
60.0% (26.2% - 87.8%) | 90.6% (79.3% - 96.9%) | 58.3% (34.5% - 78.7%) | 91.2% (82.8% - 95.7%) | |
| Cohort 3 Retrospective liver transplant evaluation |
66.7% (49.0% - 81.4%) | 80.6% (63.9% - 91.8%) | 77.4% (62.9% - 87.4%) | 70.7% (59.7% - 79.7%) | |
| Loop | Cohort 1 Retrospective compensated cirrhosis |
95.0% (75.1% - 99.8%) | 36.5% (28.1% - 45.5%) | 19.2% (16.7% - 21.9%) | 97.8% (87.0% - 99.7%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
90.0% (55.5% - 99.7%) | 81.1% (68.0% - 90.6%) | 51.1% (36.6% - 65.5%) | 97.4% (85.1% - 99.6%) | |
| Cohort 3 Retrospective liver transplant evaluation |
88.9% (73.9% - 96.9%) | 50.0% (32.9% - 67.1%) | 64.0% (55.7% - 71.5%) | 81.8% (62.8% - 92.3%) | |
| K-sparing | Cohort 1 Retrospective compensated cirrhosis |
65.0% (40.8% - 84.6%) | 46.0% (37.1% - 55.1%) | 16.0% (11.8% to 21.5%) | 89.2% (81.6% - 93.9%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
90.0% (55.5% - 99.7%) | 84.9% (72.4% - 93.2%) | 56.7% (40.1% - 71.9%) | 97.5% (85.7% - 99.6%) | |
| Cohort 3 Retrospective liver transplant evaluation |
77.8% (60.8% - 89.8%) | 58.3% (40.7% - 74.5%) | 65.1% (54.9% - 74.0%) | 72.4% (57.3% - 83.7%) | |
| Loop+ K-sparing | Cohort 1 Retrospective compensated cirrhosis |
65.0% (40.8% - 84.6%) | 56.3% (47.2% - 65.2%) | 19.1% (13.9% - 25.6%) | 91.0% (84.5% - 94.9%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
95.2% (75.5% - 99.2%) | 86.8% (74.7% - 94.5%) | 59.9% (42.1% - 75.5%) | 97.5% (86.0% - 99.6%) | |
| Cohort 3 Retrospective liver transplant evaluation |
72.2% (54.8% - 85.8%) | 61.1% (43.5%, 76.9%) | 65.0% (54.0% - 74.6%) | 68.7% (55.0% - 79.8%) | |
| Loop + K-sparing + Code | Cohort 1 Retrospective compensated cirrhosis |
55.0% (31.5% - 76.9%) | 76.2% (67.8% - 83.3%) | 26.8% (18.1% - 37.8%) | 91.4% (86.7% - 94.6%) |
| Cohort 2 Prospective Child A5-B7 cirrhosis |
60.0% (26.2% - 87.8%) | 94.3% (84.3% - 98.8%) | 69.9% (40.9% - 88.6%) | 91.5% (83.4% - 95.8%) | |
| Cohort 3 Retrospective liver transplant evaluation |
47.2% (30.4% - 64.5%) | 86.1% (70.5% - 95.3%) | 77.3% (58.4% - 89.2%) | 62.0% (53.8% - 69.5%) |
NP = negative predictive value, PPV = positive predictive value.
Performance of Identification Algorithms: Cohort 2 (Prospective 1-year follow-up)
In cohort 2, we found that loop diuretics, potassium-sparing diuretics, and a combination of both with a cirrhosis code were highly sensitive (82.3%, 82.3% and 82.3% respectively) and specific (89.1%, 93.5% and 93.0% respectively). R18.8, any ascites code and an algorithm including a combination of ascites codes and recorded diuretics were insensitive. The inclusion of a cirrhosis code did not change sensitivity (Supplementary Table 1). Meanwhile, the PPV (76.4%−84.4%) and NPV (80.8%−92.5%) were high for all the algorithms. In patients with moderate-severe ascites, a combination of recorded diuretics showed high sensitivity and specificity (95.2% and 86.8%). PPV (51.1%−77.7%) and NPV (91.2%−97.5%) for all algorithms.
Performance of Identification Algorithms: Cohort 3 (Cross-sectional at time of transplant evaluation)
In cohort 3, we found that loop diuretics, potassium-sparing diuretics, and a combination of both with a cirrhosis code were highly sensitive (90.4%, 78.8% and 75.0% respectively) and specific (85.0%, 90.0% and 95.0% respectively). R18.8, any ascites code and an algorithm including a combination of ascites codes and recorded diuretics were non-sensitive (no differences observed when including a cirrhosis code, Supplementary Table 1). Meanwhile, the PPV were high for all the algorithms (94.0%−100.0%), but NPV were lower for all of them (39.2%−77.3%) and without improvement with a cirrhosis code. In patients with moderate-severe cirrhosis, loop diuretics and R18.8 showed higher sensitivity (77.8%) and specificity (88.9%) respectively. PPV (64.0%−80.9%) and NPV (52.9%−81.8%) for all algorithms.
Discussion
To further understand the burden and impact of ascites at the population level, valid algorithms are required to identify patients in administrative data.15 Such data is lacking for ascites in the ICD-10 era. We found that recorded use of loop and potassium-sparing diuretics in patients with codes for cirrhosis, not ascites ICD-10 codes, are the optimal strategy for the identification of ascites in administrative data.
Combination Loop and potassium-sparing diuretic records optimize ascites identification
The use of loop and potassium-sparing diuretics demonstrated higher sensitivity and specificity with similar PPV and NPV compared to all diagnostic algorithms including ascites ICD-10 codes alone or in combination with recorded diuretic use. These data extend the literature in multiple ways. First, we confirm the results from Mapakshi, et al. who found that ascites ICD-10 codes (K70.31, K70.11, R18.8, and K65.2) were reliable in identifying patients with ascites with a PPV approaching 90% in the Veterans Affairs.16 Second, we extend the results from Bengtsson, et al. who found a PPV of 43% for the (infrequently used) R18.9 code for ascites, showing that it is less effective than the more commonly used R18.8.6 In both cases, neither Mapakshi nor Bengsston evaluated the NPV of the codes which we found lacking (<40% in two cohorts). Third, we find that ascites ICD-10 codes alone and combined with recorded diuretic use demonstrated suboptimal sensitivity and specificity. Instead, and most importantly, we found that recorded diuretic use demonstrated higher sensitivity and specificity for patients with ascites.
Detecting Moderate-severe Ascites in Administrative Data
A core strength of this study is the inclusion of multiple cohorts combining the spectrum of disease severity and allowing for analysis of strategies to detect moderate-severe ascites. In general, the sensitivity/NPV rose with the inclusion of diuretics for detection for all strategies but particularly for the transplant evaluation cohort (cohort 3). While the reasons for this are unclear, it is possible that as disease severity increases, coding for complications of cirrhosis may become more haphazard. The use of diuretics to express the presence of cirrhosis complications may circumvent lapses in coding.
Contextual factors
These data must be interpreted in the context of the study design. First, the performance of specific codes in hepatology and liver transplant clinics may differ against administrative data from community hospitals or primary care clinics. Bengtsson, et al. found in their study that including only university hospital data and internal medicine or transplant clinics improved the PPV of ascites ICD-10 codes.6 In addition, these results are particularly relevant in the US; however, we are unable to determine if similar results would be found in health care registers of other countries. Despite these limitations, our study has several strengths. To our knowledge, we are one of the first studies to perform a complete validation of ascites ICD-10 codes and medication records in administrative data to identify patients with ascites and moderate-severe ascites. Our multicenter study design increases the generalizability of our findings. Lastly, we evaluated ascites ICD-10 codes and recorded diuretic use with and without cirrhosis codes helping to avoid misclassification of ascites that can occur without documented underlying cirrhosis. We found that, in persons with cirrhosis, recorded use of loop and potassium-sparing diuretics in the medical record is an accurate method to identify ascites. This strategy can reliably identify patients with ascites in any database that includes pharmacy linkage.
Conclusion
These data empower the use of administrative data with medication records for the identification and study of contemporary patients with cirrhosis complicated by ascites.
Supplementary Material
4. Funding:
Elliot Tapper receives funding from the National Institutes of Health through NIDDK (1K23DK117055).
Footnotes
Disclosure:
Tapper is the guarantor of this article
Roles
a. Concept: Gonzalez, Tapper
b. Analysis: Gonzalez, Tapper
c. Data acquisition: Gonzalez, Dziwis
d. Writing: Gonzalez, Tapper, Patel
e. Revision: Gonzalez, Tapper, Patel, Dziwis
Conflicts of interest: Tapper reports grant funding from Gilead and Bausch, consulted for Kaleido, Axcella, Novo Nordisk, Novartis and Allergan, and has served on advisory boards for Bausch and Mallinckrodt. Patel has consulted for Intercept.
References
- 1.Volk ML, Tocco RS, Bazick J, Rakoski MO, Lok AS. Hospital Re-Admissions among Patients with Decompensated Cirrhosis. Am J Gastroenterol. 2012;107(2):247–252. doi: 10.1038/ajg.2011.314 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Hsu C-Y, Parikh ND, Huo T-I, Tapper EB. Comparison of Seven Noninvasive Models for Predicting Decompensation and Hospitalization in Patients with Cirrhosis. Dig Dis Sci 2021. 2021;1:1–10. doi: 10.1007/S10620-020-06763-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Piano S, Tonon M, Angeli P. Management of ascites and hepatorenal syndrome. Hepatol Int 2017 121. 2017;12(1):122–134. doi: 10.1007/S12072-017-9815-0 [DOI] [PubMed] [Google Scholar]
- 4.JL M, K S, GS W, et al. The high burden of alcoholic cirrhosis in privately insured persons in the United States. Hepatology. 2018;68(3):872–882. doi: 10.1002/HEP.29887 [DOI] [PubMed] [Google Scholar]
- 5.Fortune B, Cardenas A. Ascites, refractory ascites and hyponatremia in cirrhosis. Gastroenterol Rep. 2017;5(2):104–112. doi: 10.1093/GASTRO/GOX010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.B B, J A, JF L, H H. Validity of administrative codes associated with cirrhosis in Sweden. Scand J Gastroenterol. 2020;55(10):1205–1210. doi: 10.1080/00365521.2020.1820566 [DOI] [PubMed] [Google Scholar]
- 7.M D, E E, A H, et al. International Classification of Diseases Codes are Useful in Identifying Cirrhosis in Administrative Databases. Dig Dis Sci. June 2021. doi: 10.1007/S10620-021-07076-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Jutte DP, Roos LL, Brownell MD. Administrative Record Linkage as a Tool for Public Health Research. 10.1146/annurev-publhealth-031210-100700. 2011;32:91–108. doi: [DOI] [PubMed] [Google Scholar]
- 9.Nehra MS, Ma Y, Clark C, Amarasingham R, Rockey DC, Singal AG. Use of Administrative Claims Data for Identifying Patients with Cirrhosis. J Clin Gastroenterol. 2013;47(5):e50. doi: 10.1097/MCG.0B013E3182688D2F [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Lapointe-Shaw L, Georgie F, Carlone D, et al. Identifying cirrhosis, decompensated cirrhosis and hepatocellular carcinoma in health administrative data: A validation study. PLoS One. 2018;13(8):e0201120. doi: 10.1371/JOURNAL.PONE.0201120 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Kanwal F, Kramer JR, Buchanan P, et al. The Quality of Care Provided to Patients With Cirrhosis and Ascites in the Department of Veterans Affairs. Gastroenterology. 2012;143(1):70–77. doi: 10.1053/j.gastro.2012.03.038 [DOI] [PubMed] [Google Scholar]
- 12.Gupta K, Khan A, Goyal H, et al. Weekend admissions with ascites are associated with delayed paracentesis: A nationwide analysis of the ‘weekend effect.’ Ann Hepatol. 2020;19(5):523–529. doi: 10.1016/j.aohep.2020.05.005 [DOI] [PubMed] [Google Scholar]
- 13.Orman ES, Hayashi PH, Bataller R, Barritt AS. Paracentesis Is Associated With Reduced Mortality in Patients Hospitalized With Cirrhosis and Ascites. Clin Gastroenterol Hepatol. 2014;12(3):496–503.e1. doi: 10.1016/j.cgh.2013.08.025 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Gonzalez JJ, DiBattista J, Gomez V, et al. Impact of Inpatient Attending Specialty and Gastroenterology Consultation on Quality of Care of Patients Hospitalized with Decompensated Cirrhosis. Am J Med. June 2021. doi: 10.1016/j.amjmed.2021.05.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.EB T, S K, J B, et al. Identifying Patients With Hepatic Encephalopathy Using Administrative Data in the ICD-10 Era. Clin Gastroenterol Hepatol. 2021;19(3):604–606.e1. doi: 10.1016/J.CGH.2019.12.017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.S M, JR K, P R, HB E-S, F K. Positive Predictive Value of International Classification of Diseases, 10th Revision, Codes for Cirrhosis and Its Related Complications. Clin Gastroenterol Hepatol. 2018;16(10):1677–1678. doi: 10.1016/J.CGH.2018.01.042 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.