Propensity-score-matched evaluation of under-recognition of acute kidney injury and short-term outcomes

Acute kidney injury (AKI) is a common disease, but diagnosis is usually delayed or missed in hospitalized patients. The aim of this study was to investigate the impact of under-recognition of AKI (beyond 3 days after AKI onset) on short-time prognosis. Of 785 patients with under-recognition of AKI and 616 patients with timely-recognition of AKI were propensity matched in a 1:1 ratio. The two groups, with a total of 482 matched patients (241:241), were comparable in baseline covariates. Under-recognition of AKI was not associated with 30-day all-cause mortality in the logistic regression model with covariate adjustment (OR = 1.01, 95% CI = 0.62–1.64, p = 0.967). Sensitivity analyses and subgroup analyses also proved the association. There were also no significant differences in causes of 30-day mortality, in-hospital mortality, recovery of renal function at discharge, length of hospital stay, length of intensive care unit stay or hospitalization costs between the two groups, although timely-recognition group had more chance of renal consult and a little more interventions for AKI. In conclusion, under-recognition of AKI may not be associated with poor short-term outcomes of adult hospitalized patients via these propensity-score-matched analyses.

Definitions. The baseline SCr value was set based on the lowest measurement (≥40 μmol/L) during hospitalization 16 . AKI was defined according to the KDIGO criteria as an increase in SCr to ≥26.5 μmol/L within 48 h or ≥1.5 times the baseline value either known or presumed to have occurred within the prior 7 days 17 .
Complete recovery of renal function was defined as the recovery of SCr to no more than the baseline value of 44.2 μmol/L and partial recovery was defined as greater than the baseline value of 44.2 μmol/L, but no more than the maximum value and discontinuing dialysis >1 week at the time of discharge. Failure to recovery was defined as continuous SCr elevation or requirement of maintenance dialysis 18 .
Timely recognition of AKI was defined as diagnosis within 3 days after onset, as documented in the electronic medical record system 8 . Under-recognition of AKI was defined as either a delayed or missed diagnosis. Delayed diagnosis of AKI was defined as recording AKI more than 4 days after onset 8 . Missed diagnosis of AKI was defined as no relevant diagnostic record in the medical record system. In this study, sepsis was defined as infection-induced systemic inflammatory response syndrome according to the 1992 definition 19 . Surgery included elective surgery, time-limited surgery, emergency surgery, and other surgical treatments.
Outcomes evaluation. The primary clinical endpoint was 30-day all-cause mortality. Secondary clinical outcomes included in-hospital mortality, recovery of renal function at discharge, average length of hospital stay, average length of ICU stay, and hospitalization costs.
Statistical methods. All data analyses were conducted using SAS version 9.2 statistical software.
Continuous variables with a normal distribution were presented as the mean (±standard deviation) and compared between groups using the Student's t-test. Continuous variables with a skewed distribution were presented as the median (range) and compared between groups using the rank-sum test. Categorical variables were expressed as the number (percentage) and compared between groups using the chi-squared test.
A multivariate logistic regression model was used to yield a PS for each patient. The dependent variable was whether AKI recognition was timely or not. There were 17 independent variables, including sex, age, department that AKI occurred, CCI, eGFR on admission, oliguria, history of CKD, history of malignant neoplasms, APACHE II score, SOFA score, risk factors for AKI (hypovolemia, heart failure, sepsis, surgery), AKI stage, blood urea nitrogen, and whether RRT was received. The PS was used to produce a 1:1 match, which was the closest with no replacement. The caliper was 0.2 times the root mean square value of the standard deviation of the two groups (0.05 after calculation) 20 . Model discrimination was assessed with c statistics (0.913), and model calibration was assessed with Hosmer-Lemeshow statistics (χ 2 = 12.150; df = 8; p = 0.145).
After all PS matches were performed, differences in patient characteristics between the 2 groups were compared by standardized differences, of which 0.8, 0.5, and 0.2 were considered large, medium, and small differences, respectively, and ≥0.1 was defined as meaningful imbalance 21,22 . Survival curves were generated using the Kaplan-Meier method and compared using the log-rank test. To reduce the influence of possible confounding variables, we used the PS matched analysis, inverse probability of treatment weighting (IPTW) and stratification based on quintiles of the PS scores. In the PS-matched cohort, the risks of mortality between groups were compared with logistic regression using Generalized Estimating Equations. A two-sided p value of less than 0.05 was considered statistically significant.

Results
Baseline characteristics of patients. From October 1, 2013 to September 30, 2014, a total of 87,196 patients were admitted to the first Affiliated Hospital of Nanjing Medical University. Of these patients, 26,869 (30.8%) had two or more SCr values during hospitalization and 621 were excluded: 468 with end-stage renal disease, 115 with kidney transplant, 30 aged <18 years, and eight who were hospitalized for <24 h. Finally, 1401 cases (1.61%) in this cohort were diagnosed with AKI ( Fig. 1).
According to whether AKI was timely recognized, the patients were divided into two groups: a timely recognition (TR) group (616 cases) and an under-recognition (UR) group (785 cases). There were significant differences between groups in patient age, department distribution, complications (lung disease, malignancy, and CKD), CCI, AKI risk factors (usage of contrast agent, chemotherapeutic agent, heart failure, sepsis, surgery), AKI stage, and laboratory parameters (hemoglobin, platelet, albumin, total bilirubin, urea nitrogen, and eGFR) ( Table 1).
After propensity score matching, 241 cases were assigned to each group (Fig. 1). After matching, an adequate comparability was shown by a decrease to less than 0.2 of the standardized difference between TR group and UR group for all baseline matched covariates (Table 1). Apart from diuretic usage, there were also no significant differences in other unmatched baselines covariates including reasons for admission, comorbidities, risk factors of AKI and laboratory data on admission (i.e., measurements of white blood cells, hemoglobin, platelets, albumin and serum total bilirubin) ( Table 1).  Table 2). The Kaplan-Meier plots demonstrated that there was no difference in 30-day mortality between these two groups (p = 0.794, Fig. 2). Furthermore, there was no difference in the causes of 30-day mortality (p = 0.903, Table S1) or any secondary endpoint (i.e., in-hospital mortality, renal function recovery when discharged, average length of hospital stay, average length of ICU stay, hospitalization costs, and daily hospitalization cost) after matching between the TR and UR groups. The UR group was also not associated with 30-day mortality in univariate and multivariate logistic regression analysis using Generalized Estimating Equations (Tables S2-S3).

Interventions in the AKI patients.
Patients in the TR group had greater numbers of renal consults within 3 and 7 days, a higher rate of stopping ACEI/ARB within 7 days following AKI, and higher rate of renal ultrasound examination within 3 days (Table 3). However, there were no differences between the matched groups in stopping nephrotoxic antibiotics within 3 days, use of non-steroidal anti-inflammatory drugs (NSAIDs), use of contrast agents, transfusion within 3 days, urinalysis within 3 days, and SCr measurement within 3 days.

Sensitivity analyses.
The association between under-recognition of AKI and 30-day mortality were indicated in Table 4. PS matching using either crude estimate (OR = 1.04, 95% CI = 0.84-1.29, P = 0.711) or covariate adjustment (OR = 1.01, 95% CI = 0.62-1.64, P = 0.967) showed that under-recognition of AKI did not associated with higher 30-day mortality. Due to inherent drawback that reduced sample sizes exist in PS matching, traditional logistic regression, PS as covariate, PS stratification and ITPW were also used to calculate the association in these sensitivity analyses 23 . The univariate logistic regression in overall cohort showed that under-recognition of AKI was associated with reduced 30-day mortality (OR = 0.36, 95% = 0.28-0.45, P < 0.001). However, the association became statistically insignificant (OR = 0.89, 95% = 0.60-1.32, P = 0.575) after adjustment of 17 matched covariates. In addition, the overall estimated treatment effect from stratification based on PS (Table S4), PS as covariate and PS as covariate "doubly robust" also proved this association. In ITPW models, the association became statistically insignificant after adjustment of 17 matched covariates in Table 4.

Subgroup analyses.
Subgroup analyses according to stratification of age, sex, AKI stage, APACHE II score, and department distribution, types of AKI and primary reasons for hospital admission showed that failure to timely recognize AKI did not significantly increase 30-day all-cause mortality in all subgroups in the matching  cohort (Fig. 3). In addition, PS matching in specific population with distribution in internal medicine/surgery/ ICU department, or with varying types of AKI and primary reasons for hospital admission also proved this result (Table S5).

Discussion
In this study, the effect of delayed or missed recognition (under-recognition) of AKI on short-term outcomes of adult hospitalized patients was investigated. Using a propensity score matching method, under-recognition of AKI was not associated with a poorer short-term outcome, which was confirmed by the sensitivity analyses and subgroup analyses. AKI is an important healthcare burden worldwide 24 , but under-recognition of AKI is very common, which may delay the prompt adoption of protective interventions, including volume repletion, blood pressure optimization, hematocrit correction, maintenance of adequate oxygen saturation, and avoidance or discontinuation of nephrotoxic drugs. Therefore, under-recognition of AKI was assumed to increase mortality or limit the recovery of renal function in AKI patients. But, the evidence supporting this viewpoint remains insufficient. In 2008, Wilson et al. 7 found that missed recognition of AKI could increase 30-day mortality by logistic regression stepwise correction, in which the correction factor was the largest SOFA score over the course of the disease. The SOFA score itself could be affected by misdiagnosis of AKI. Thus, this research could not fully confirm the above assumption. Besides, a randomized controlled trial to detect the effect of early recognition of AKI on short-term prognosis is not feasible in the real clinical world. Therefore, this propensity score matching study was designed to preliminarily discuss the issue.
The results of this study showed that under-recognition of AKI did not significantly increase the risk of short-term adverse outcomes, including 30-day all-cause mortality, in-hospital mortality, and recovery of renal function at discharge. There are several possible explanations for these results. First, delayed or missed diagnosis of AKI actually may not be associated with poorer prognosis because of the lack of specific effective treatment for AKI, as prevention may be more important than treatment. A recent randomized controlled study reported that increasing the diagnostic rate of AKI through the use of an electronic warning system did not improve prognosis,   which also supported our results 25 . Second, increased recognition rates (such as from 5% to 40%) could lead to improved prognosis by greatly improving the effective treatment of AKI, but when the diagnostic rate increased to a certain level, such as from 40% to 60%, timely diagnosis may lead to little improvement in mortality because available treatment options to further improve prognosis were limited and difficult. Third, timely recognition did not mean very early recognition of AKI, which required physicians to recognize AKI within several hours. Perhaps the use of a more sensitive indicator for very early recognition of AKI could improve outcomes. Lastly, some results may be false-negatives because the median sample size of this study resulted in insufficient power (β = 0.067) to prove these findings. Thus, a larger AKI population is needed to confirm this result. Early recognition of AKI without adequate intervention did not improve short-term outcomes. Our study showed that the chance of renal consult, the rate of stopping ACEIs or ARBs, the rate of renal ultrasound examination, and the use of diuretics increased in the timely-recognition group as compared to the under-recognition group, although diuretics were not recommended in all cases of AKI with the exception of fluid overload 26 . In addition, our results showed that there were no differences between the matched groups in stopping nephrotoxic antibiotics within 3 days, use of NSAIDs, use of contrast agents, and transfusion within 3 days, which may attribute to the inadequate knowledge and training systems and insufficient multidisciplinary cooperation 27 . Obviously, these factors were not sufficient to improve the short-term outcomes of AKI patients. There was also an increasing trend in the use of RRT in the TR group as compared to the UR group. A recent randomized study showed that an automated, electronic alert system for AKI did not improve outcomes and interventions except frequency of dialysis in surgical patients 25 , which was similar to our results. This indicated that early recognition of AKI without adequate intervention could not improve short-term outcomes. Therefore, adequate intervention is as important as early recognition of AKI.   There were several limitations to this study that should be addressed. First, the sample size of this study (n = 482) resulted in insufficient power (β = 0.067). If the difference in the survival rate was 1.7% and α was set to 0.05 and β to 0.90, then 17542 cases would be needed in each group to detect a difference in survival rates. Thus, future studies with larger cohorts are required to verify this hypothesis. Second, the study population was distributed among different departments, so the primary disease, aggravating factors, and risk factors were heterogeneous. The matching method in this study may not be perfect despite the inclusion of 17 factors. Third, short-term, rather than long-term, prognosis was observed in this study because even stage 1 AKI could increase the long-term risk of CKD, and AKI patients with delayed or missed recognition are more likely to develop advanced CKD if there is no long-term follow-up of renal function. Hence, future research of the beneficial effect of improving AKI recognition on long-term outcomes is required.

Conclusion
The results of this propensity score matched study showed that under-recognition of AKI may not be associated with adverse short-term outcomes in hospitalized adult patients, which may attribute to the inadequate interventions for AKI. Due to the limitation of the sample size, further studies are required to confirm this result. Future studies are also warranted to investigate the effect of under-recognition of AKI on long-term outcomes.

Data Availability
The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.