Outpatient acute kidney disease detection by national and institutional health data

Table of Contents

Epidemiology and clinical characteristics of AKD_OPT

Between December 2017 and May 2020, the monthly cumulative incidence of AKD_OPT was stable among eligible population, with the average incidence being 11.0% (95% confidence interval [CI] 10.8–11.2%), and the top 3 clinical divisions that had the highest number of AKD_OPT requiring nephrology referral were cardiology, urology, and gastroenterology (Fig. 2; Supplementary Figs. 2, 3). Moreover, 23.0% (95% CI 21.9–24.0%) of the patients with AKD_OPT were considered to be high risk and required nephrology referral, but only 2.8% (95% CI 2.6–2.9%) of the patients with AKD_OPT actually had a nephrology clinic follow-up in the same month. In the sensitivity analyses of limiting eligible outpatient visits with at both diagnostic S-Cre measurements from CMUH’s outpatient setting or with at least one S-Cre measurements from CMUH’s outpatient, the average incidence of AKD_OPT decreased to 2.9% (95% CI, 2.8–3.0%) and 6.1% (95% CI 5.9–6.3%), respectively. However, the proportion of high-risk AKD_OPT remained high (32.3% [30.4–34.2%] and 28.5% [27.1–29.9%]) and the rate of AKD_OPT patients having nephrology follow-up remained low (4.3% [3.8–4.8%] and 3.5% [3.3–3.7%]; Supplementary Figs. 4, 5).

**Fig. 2: Epidemiology profile of AKD_OPT from December 2017 to May 2020 among eligible adult outpatients free from nephrology care, ESKD, and cancer.**

Of the 79,838 eligible outpatients (study population), 12,510 (15.7%) were identified by the AKIDS as having AKD_OPT, with 81.0% of them having stable AKD_OPT and 19.0% having deteriorating AKD_OPT. Compared with patients without AKD_OPT, those with AKD_OPT were older, were more likely to be a former or active tobacco smoker or alcohol drinker, and had a higher burden of comorbidities such as diabetes and hypertension (Table 1). Patients with AKD_OPT were more likely to have an advanced CKD stage (eGFR <60 mL/min/1.73 m²) (18.3% vs 10.5%), had higher blood urea nitrogen and pooled urine albumin-to-creatinine ratio levels but lower serum albumin and hemoglobin levels than did those without AKD_OPT (Table 1 and Supplementary Data 4). Because the setting of the S-Cre measurements used to screen for AKD_OPT could only be verified if the measurements were from the on-site electronic medical records and because the inclusion of S-Cre from inpatient measurements may interfere the AKD_OPT detection, we provided the distribution of S-Cre source. Only 45.2% of the study population had their maximum and minimum S-Cre measurements solely from the outpatient setting, 20.9% from inpatient and 34% from other institutions via the NHI Health Information Exchange (Table 1). The patients with AKD_OPT had significantly higher 1-year CKO (12.4% vs. 0.5%), all-cause mortality (8.2% vs. 1.4%), and de novo CKD-ND rates (7.8% vs. 1.6%) than did those without AKD_OPT.

Table 1 Baseline demographic and clinical characteristics of the AKD_OPT study population selected from CMUH for the period between December 2017 and June 2019

Associations of AKD_OPT with CKO and all-cause mortality

During the 1-year follow-up, 1942 patients died and 1858 had CKO (96 had ESKD; 1702 had a > 40% decrease in eGFR; 783 had a twofold increase in S-Cre levels; not mutually exclusive). The incidence of 1-year CKO was notably higher in patients with AKD_OPT than in those without AKD_OPT (12.77 vs 0.42 per 1000 person-months); the adjusted absolute risk was 79 (95% CI 78.9–79.1) per 1000 persons. The corresponding difference for deteriorating AKD_OPT was even higher, reaching 109.7 (95% CI 109.5–109.9) per 1000 persons (Table 2). The 1-year mortality rate was significantly higher in patients with AKD_OPT than in those without AKD_OPT (7.12 vs 1.15 per 1000 person-months); the adjusted absolute risk was 25.3 (95% CI 25.2–25.3) per 1000 persons. The absolute risk of mortality was higher in patients with stable AKD_OPT than in those with deteriorating AKD_OPT. Moreover, the aHRs for 1-year CKO and mortality in the AKD_OPT group relative to the non-AKD_OPT group were 16.2 (14.2–18.5) and 2.6 (2.4–2.9), respectively (Table 2). The aHRs for 1-year CKO and mortality in the deteriorating-AKD_OPT group relative to the non-AKD_OPT group were 22.7 (19.5–26.4) and 2.1 (1.8–2.6), respectively; the same aHRs in the stable-AKD_OPT group relative to the non-AKD_OPT group were 14.3 (12.5–16.4) and 2.8 (2.5–3.1), respectively. Among patients without baseline CKD, the corresponding aHRs for 1-year de novo CKD-ND in the AKD_OPT, deteriorating-AKD_OPT, and stable-AKD_OPT groups relative to the non-AKD_OPT group were 3.5 (3.1–3.9), 4.1 (3.5–4.8), and 3.3 (2.9–3.7), respectively. For each outcome, the time-to-event curves were significant different between AKD_OPT and non-AKD_OPT groups, as well as across AKD_OPT subgroups (Fig. 3a–c and Supplementary Fig. 6a–c).

Table 2 Hazard ratios, absolute risk, and their corresponding 95% confidence intervals for 1-year CKOs, all-cause mortality, and de novo CKD associated with AKD_OPT and its subtypes (data with multiple imputation)

**Fig. 3: Risk of outcomes associated with AKD_OPT.**

Of patients who were younger than 65 years or those were free from comorbidities such as diabetes, hypertension, or CKD at baseline, having AKD_OPT was associated with a significantly higher risk of 1-year CKO, mortality, and de novo CKD-ND (Supplementary Fig. 7). The risk matrix demonstrated that the aHR for each outcome exhibited a gradient increase from low-risk (non-AKD_OPT and ultimate eGFR of ≥60) to high-risk patients (AKD_OPT and ultimate eGFR of <45) (Fig. 3d–f and Supplementary Fig. 6d–f).

We conducted various sensitivity analyses to ensure the robustness of our results. The aHR and absolute risks from these analyses aligned with our main findings under different conditions: (1) without multiple imputation (Supplementary Table 4), (2) with diagnostic S-Cre measurements exclusively from CMUH’s outpatients (Supplementary Table 5) and with at least one diagnostic S-Cre from CMUH’s outpatients (Supplementary Table 6), () with at least one outpatient S-Cre during follow-up (Supplementary Table 7), (4) with diagnostic S-Cre taken within the standard 90-day AKD diagnostic window (Supplementary Table 8), and (5) with diagnostic S-Cre taken from CMUH’s outpatients within 90-day window (Supplementary Table 9). By applying the most stringent definition for AKD_OPT, we found larger effect sizes for the aHRs and adjusted absolute risks (Supplementary Table 9), which implies that our original definition might have underestimated the clinical impact of AKD_OPT. For the 87,426 patients initially excluded because of an unknown AKD_OPT status, we adopted the CMUH’s median outpatient S-Cre levels (adjusted for age, sex, and CKD status) as the baseline S-Cre. The association of AKD_OPT with 1-year CKO, mortality, and de novo CKD remained significant, albeit with a reduced effect size (Supplementary Table 10).

Recovery pattern of AKD_OPT and nephrology follow-up

Of the patients with AKD_OPT, 23.4%, 70.0%, and 41.6% experienced recovery to their baseline S-Cre levels (absolute recovery), recovery to ≤0.2 mg/dL above baseline levels, and recovery to ≤1.1 times the baseline levels within 1 year, respectively (Table 1). However, compared with the patients with stable AKD_OPT (22.4%), those with deteriorating AKD_OPT had a higher absolute recovery rate (27.1%). This may be attributed to the fact that patients with deteriorating AKD_OPT were more likely to be regularly followed by nephrologists (12.2% vs 7.0%) and were follow-up sooner (23 days vs 48 days), compared to those with stable AKD_OPT. Among the patients with CKD at baseline, 22.4%, 19.3%, and 29.4% of the patients with AKD_OPT (all), stable AKD_OPT, and deteriorating AKD_OPT had CKD stage progression, respectively (Table 1). The nephrology follow-up rate within 90 days following the index date was 2.7%, 2.2%, and 4.9% for the patients with AKD_OPT, stable AKD_OPT, and deteriorating AKD_OPT, respectively.