Uncoded chronic kidney disease prevalence in secondary care: a retrospective audit with population health implications

Retrospective audit of inpatient admission data from 2nd April 2022 to 22nd February 2023 as part of a Population Health Fellowship in Chronic Kidney Disease and University of Manchester Medical School– medical student Applied Personal Excellence Project. Primary aims: to measure total and coded/uncoded CKD prevalence on admission and discharge, and conversion of uncoded to coded CKD in secondary care. Secondary aims: to map coding status to kidney health inequality themes and to measure predictors of coding, death and AKI. Diagnosis and coding of CKD are key standards within NICE CKD guidelines [1] and the primary care Quality and Outcomes Framework (QOF) [7], and are further supported by expert consensus [12].

All adult patients admitted to the 21-bed Clinical Assessment Unit in Rochdale Care Organisation (RCO), including multiple admissions by the same patients, were included. Inpatient admission data were extracted from the electronic health record (EHR)– HealthViews.

Ethical approval was not required for this audit as per UK Health Research Authority (HRA) guidance which was registered with and endorsed by the Northern Care Alliance (NCA) National Health Service (NHS) Foundation Trust Research and Innovation department (project code: 23HIP17).

The following outcomes were used: prevalence estimates, calculated as the proportion of patients with the outcome of interest (numerator) divided by the total number of patients in the relevant group (denominator), expressed as a percentage; coded CKD based on admission or discharge eGFR and latest uACR in preceding 12 months; conversion from uncoded to coded CKD at discharge; odds of patients receiving a diagnostic CKD code on discharge, death during admission, and AKI during admission (defined according to KDIGO definition [13]).

Demographic and predictor variables included: age and age groups (17–29, 30–39, 40–49, 50–59, 60 − 59, 70–79, 80–89, 90 + years), sex, ethnicity, GP practice post code– mapped to indices of multiple deprivation (IMD) [14]: a measure of geographical area level deprivation at a low geographical level of approximately 1600 people, measured over several domains (income; employment; education, skills, and training; health deprivation and disability; crime; and housing) - used as a proxy for socioeconomic status; admission and discharge date; first and last creatinine on admission (µmol); uACR (mg/mmol); past diagnostic codes in primary care: CKD, diabetes, hypertension, mental health diagnoses (anxiety, depression, bipolar disorder, schizophrenia and psychosis, self-harm and suicidal ideation, eating disorder, dementia), cancer; diagnoses of AKI and/or heart failure during admission; death during admission. Derived variables included eGFR (ml/min/1.73m²) using the CKD-EPI 2021 equation [15]. CKD was defined and staged according to the KDIGO classification criteria using 2 x eGFR values within 90 days and the latest uACR in the preceding 12 months (if measured) [3]; patients with eGFR > 60 ml/min/1.73m² without uACR were classified as ‘Not CKD’. Admission eGFR was the first eGFR during admission. Discharge eGFR was the last eGFR during admission. For patients with only one eGFR measured on admission, this was carried forwards as their discharge eGFR. Coded CKD was defined as any patient with biochemical evidence of CKD who either had a diagnostic CKD code in their primary care EHR or who were given a diagnostic CKD code on discharge from hospital. Patients with coded CKD were accepted as having CKD and this diagnosis was not challenged. Uncoded CKD was defined as any patient with biochemical evidence of CKD on discharge without ever having a diagnostic CKD code in their primary care EHR. Kidney health inequality themes are overarching themes which highlight patterns of disparities in kidney health outcomes for patients. Coding status was mapped to kidney health inequality themes of sex, age, ethnicity, socioeconomic status, and mental health diagnoses [10].

To minimise selection and information bias we included all admitted patients within the prespecified study period and cross-checked hospital level data with primary care EHR data to accurately identify patients with uncoded and coded CKD.

R and R Studio were used for data cleaning and analysis, employing packages dplyr, tidyr, logistf, patchwork and ggplot2. Descriptive statistics (counts/percentages for categorical data; mean and standard deviation for continuous data) were reported. Demographics were compared between the total cohort and patients with CKD. Firth’s penalised likelihood method was used for logistic regression analyses to address the sample size limitations and potential separation [16, 17]. Analyses assessed key predictors of diagnostic coding at discharge, in-hospital mortality, and AKI risk during admission. An alpha level of 0.05 was set a priori and 95% confidence intervals (CIs) are presented. Logistic regression assumptions were verified before modelling. Likelihood ratio tests and p-values guided model selection.

Of 482 patients with CKD, 199 (41.3%) were coded in primary care on admission and a further 283 (58.7%) had evidence of CKD using discharge eGFR and uACR but were uncoded in primary care. Of patients with uncoded CKD (n = 283), only 19 were coded on discharge, representing a 6.7% conversion of uncoded to coded CKD, leaving 264 patients with uncoded CKD. Figure 1 shows the prevalence of uncoded CKD generally decreases as CKD stage increases. Uncoded CKD was more common up until stage G4.

Figure 2 highlights coding status across 5 kidney health inequality themes. Age: coded CKD was more common as age increases. Sex: Less than half of males or females had coded CKD. Ethnicity: Less than half of all ethnicities had coded CKD. Socioeconomic status: in general, coded CKD becomes more common as deprivation decreases. Mental health: coded CKD is more common in patients with 1 or more MH diagnoes. Overall, across health inequality themes, uncoded CKD was generally more common than coded CKD.

Of 482 patients with CKD, 168 (34.9%) had diabetes; uACR was measured in 123 (73.2%). Figure 3a shows there were no patients with diabetes and coded CKD stage 1, and uncoded CKD was more common until stage 4. There were no patients with diabetes and uncoded CKD stage 5.

Of 482 patients with CKD, 280 (58.1%) had hypertension; uACR was measured in 173 (61.8%). Figure 3b shows coded CKD generally becomes more common as CKD stage increases. There were no patients with hypertension and uncoded CKD stage 5.

In patients with CKD, age (OR 1.07; CIs 1.01–1.16), sex (OR 0.27; CIs 0.05-1.00) IMD decile (OR 0.62; CIs 0.32–0.94) and AKI during admission (OR 3.92; CIs 1.14–14.82) were all significant predictors of death during admission (supplement 8.1: final model).

In patients with CKD, advanced CKD (OR 3.08; CIs 1.96–4.88) was the only significant predictor of AKI during admission (supplement 8.2: final model).

There is scant evidence of measurement of CKD prevalence in hospitalised patients. Our study demonstrates a greater prevalence of CKD in acute medical inpatients (35.3%) than existing observational research: 14.8% of 13,383 adult inpatients in China [20]; and 12.7% in 826 adult acute medical inpatients in Brazil [21] which may be a function of our study population being older, living in greater levels of deprivation, with different access to healthcare, and in an acute medical ward. Such a finding aligns with kidney health inequality data showing that older adults not only face barriers to diagnosis but also effective CKD care [11]. Nevertheless, earlier stages of CKD are coded less often compared to later stages of CKD, a pattern observed in existing research [2, 4]. This discrepancy arises from factors including asymptomatic earlier stages of CKD, variable clinical recognition, limited screening, and competing demands in primary care [2, 8, 22]. Existing evidence focuses on primary care as a central setting for improved detection, diagnosis and coding of CKD [2]; however, our findings highlight not only the high prevalence of uncoded CKD in a hospital setting but therefore the opportunity for intervention in an additional setting to primary care. This aligns with the national strategy for CKD care highlighting the role of all healthcare professionals in addressing kidney health inequalities [11]. Furthermore, coding a diagnosis of CKD increases the chances of optimal management and better health outcomes [4, 5]. We show that a diagnosis of hypertension (OR 3.14; CIs 1.14–9.97) and advanced CKD (OR 4.90; CIs 1.60–14.20) significantly increase the odds of patients with uncoded CKD receiving a diagnostic CKD code on discharge. Whilst these predictors have wide CIs indicating some uncertainty in the exact estimates, such predictors are clinically sound and correlate with existing evidence [2, 4]. Observational research also shows uncoded CKD in a hospitalised older adult population was 50.8% and associated with higher inappropriate prescribing of medications [23] emphasising the clinical implications of uncoded CKD in a hospital setting and for a vulnerable patient population. Furthermore, patients with uncoded CKD may be at greater risk of having renoprotective medication stopped and not restarted. In our study, uncoded CKD was generally more prevalent than coded CKD across kidney health inequality themes demonstrating the potential for coding as an intervention to have equity across themes. We show uACR testing is suboptimal and below guideline recommendations for high-risk patients (diabetes/hypertension) [1]. Our findings also contribute to the overall limited evidence base of mental health disease prevalence in patients with CKD [10, 24‐27].

Undiagnosed and uncoded CKD is often considered a primary care problem. However, there is a notable gap in the evidence examining this issue within hospital settings. Our study addresses this gap by thoroughly auditing inpatient admissions and correlating these findings with primary care EHRs diagnoses. This approach allows for a comprehensive understanding of a patient’s health status, presenting a new view of the problem of uncoded CKD from another vantage point. Additionally, our study aligns findings relating to coding status with the latest evidence on kidney health inequalities to add further colour to this picture. To our knowledge we present the first evidence on conversion rates of uncoded to coded CKD in a hospital setting which sets the benchmark for future research. Limitations include that eGFR fluctuates during acute illness which may overestimate CKD stage. IMD was calculated on GP practice postcode, not patient postcode; GP practices were heavily based in areas of IMD deciles 1–3 where uncoded CKD prevalence is higher, potentially limiting generalisability to other populations. Missing uACR underestimates earlier stages of CKD and risk of poor outcomes. uACR was included from within the previous 12 months, not necessarily at the time of eGFR testing during the hospital admission. All analyses were of routinely collected data which underreport true CKD prevalence. Furthermore, the retrospective nature of our study carries a risk of residual confounding.