Skip to content
PaperFren

Can billing codes tell us who has chronic kidney disease?

Open paper intelligence

Hospital and doctor billing codes rarely flag kidney disease falsely, but they miss about two out of three older people whose blood tests show reduced kidney function.

Source

Detecting chronic kidney disease in population-based administrative databases using an algorithm of hospital encounter and physician claim codes

Fleet JL, Dixon SN, Shariff SZ, et al. · BMC nephrology · 2013

doi.org/10.1186/1471-2369-14-81Read the full paper ↗145 citationscc by

Study at a glance

Design
Cross-sectional — Retrospective diagnostic-validation study: presence of any of 11 CKD codes in the prior five years (index test) compared against the most recent laboratory eGFR (reference standard) in linked Ontario databases.
N
N=123499 · 123,499 adults aged 66 or older in Southwestern Ontario who filled an outpatient prescription between 2007 and 2010 and had a serum creatinine test in the prior year.
Population
Older adults (66+) in Southwestern Ontario, Canada, with universal health coverage, excluding those on dialysis or with a kidney transplant.
Outcome
Sensitivity, specificity, positive and negative predictive value of the coding algorithm for detecting eGFR below 45 (and below 60 and 30) mL/min per 1.73 m2.

Structured fields used in claim comparison tables when every cited study has a complete layer.

What they did

Researchers in Ontario linked provincial health databases to laboratory results for 123,499 adults aged 66 and over. Two nephrologists picked 55 possibly relevant diagnosis and billing codes, which were narrowed to a final algorithm of 11 codes; a person counted as 'code positive' if any of these appeared at least once in the previous five years. They compared code status with each person's most recent estimated glomerular filtration rate (eGFR), treating an eGFR below 45 as the reference definition of CKD.

What they found

About 15% of the cohort had an eGFR below 45, but only 7.7% were flagged by the codes. The algorithm's sensitivity was 32.7% and its specificity 96.9%; positive predictive value was 65.4% and negative predictive value 88.8%. Sensitivity was better for more severe disease (58.8% for eGFR below 30) and was lower in women than men (25.7% vs 43.7%) and in people over 80 than in those aged 66 to 80.

The limits

What it doesn't show

CKD was defined from a single eGFR value, whereas clinical diagnosis needs two measurements at least three months apart, and CKD defined by protein in the urine was not captured. The algorithm was built and tested on the same data with no separate validation sample, so its performance may be optimistic and may not transfer to younger people, other provinces or systems without fee-for-service billing. Codes were combined only with a simple 'any code' rule. Because sensitivity differs by sex and age, studies using these codes could produce biased comparisons between groups.

Key terms

Sensitivity
The share of people who truly have the condition (by the reference standard) that the test correctly flags.
Specificity
The share of people who truly do not have the condition that the test correctly leaves unflagged.
Positive predictive value (PPV)
Among people the test flags, the proportion who actually have the condition; it depends on how common the condition is.
Reference standard
The best available measure of the true condition against which a new test is judged; here, laboratory eGFR.
eGFR
Estimated glomerular filtration rate, calculated from blood creatinine (here using the CKD-EPI equation) to estimate how well the kidneys filter blood.
Administrative database
Records collected for running the health system, such as hospital discharges and physician billing claims, rather than for research.

Flashcards

1 / 11

0 of 11 answers reviewed

Research intelligence for this paper

See its role on concept claims, tensions it is part of, placement history, and related discoveries.

Open paper intelligence

Quiz yourself

1 / 6

What was the approximate sensitivity of the final code algorithm for detecting an eGFR below 45?

Common questions

If sensitivity is only about a third, is the algorithm useless?

Not entirely. Its high specificity and negative predictive value mean that people it does flag usually have reduced kidney function, so it can serve as a baseline characteristic or adjustment variable, but it badly undercounts CKD and should not be used to estimate how common CKD is or as a main outcome.

Why might women and the very old be missed more often?

Codes only appear when a doctor records a diagnosis. Mild or age-related declines in kidney function, and CKD in women (who have lower creatinine for the same kidney function), are more often unrecognised, so they never get coded.

Why use an eGFR cut-off of 45 rather than 60?

In older people an eGFR below 45 is widely accepted as clinically important and is more stable on repeat testing, whereas values between 45 and 60 are common in the elderly and more debated.

More on Kidney disease