Risk of Hospital-Acquired Acute Kidney Injury among Adult Opioid Analgesic Users: A Multicenter Real-World Data Analysis

Acute kidney injury (AKI), characterized by a sharp decline in renal function over a short period, is common in hospitalized adults [1‒3] and associated with substantial adverse outcomes [3‒5]. Exposure to nephrotoxic medications is a major cause of hospital-acquired (HA) AKI, contributing to 14%–26% of all cases [6]. Therefore, nephrotoxicity should be taken into consideration when choosing the drugs.

Proper use of opioid analgesics, a class of medication that acts on opioid receptors, is a safe and effective method for pain relief [7]. The global use of opioid analgesics has increased dramatically in recent decades, doubling from 3 billion defined daily doses in 2001 to 7.3 billion defined daily doses in 2011 [8]. A similar increasing trend was reported in a national survey in China [9]. There are many opioid analgesic drugs on the market, including natural alkaloids (morphine, codeine) [10]; semi-synthetic derivatives (oxycodone, dihydrocodeine, and hydromorphone) [11, 12]; and synthetic derivatives (fentanyl transdermal patch, pethidine, tramadol, pentazocine, dezocine, butorphanol, and nalbuphine) [10, 12, 13]. However, the nephrotoxicity of these drugs has not been extensively studied in patients. Two recent studies suggested that morphine use was associated with an increased risk of AKI [14, 15], though these two studies were limited by small sample sizes. For opioid analgesics other than morphine, there were only a few sporadic case reports for possible nephrotoxicity [16‒20]. Most of the AKI events in these case reports occurred on the day or within a few days of opioid analgesic overdoses, and possible mechanisms include hypotension and rhabdomyolysis [19, 20]. In addition, adverse effects of opioids include nausea and urinary retention, which may also contribute to the development of AKI [21]. We conducted a large multicenter, retrospective real-world study to systematically compare the risk of HA-AKI among 255,265 adult users of opioid analgesics.

The retrospective real-world study cohort was drawn from the China Renal Data System (CRDS). The CRDS is a cooperative network formed by the regional medical centers across China with the aim of facilitating clinical research on kidney disease. Each participating center exported from its proprietary hospital information systems the demographic and clinical data of the patients who had a history of hospitalization between 2000 and 2021. The exported data were cleaned up, standardized, anonymized, and pooled at the CRDS datacenter located at the National Clinical Research Center of Kidney Disease in Guangzhou. The demographic data included gender and date of birth. The clinical data included admission and discharge date and division of each hospitalization, diagnosis codes, operation and procedure codes, results of laboratory assays, imaging and histological reports, prescriptions, and medical records of both in-patients and out-patients visits. All of the laboratories of the participating centers had passed the annual External Quality Assessment by the Chinese National Center for Clinical Laboratories. As of November 1, 2021, the CRDS databases included the data of more than 7 million patients from 19 medical centers. The standardization processes in the CRDS database and the CRDS investigators are shown in online supplementary methods (for all online suppl. material, see https://doi.org/10.1159/000533556 for details). The Medical Ethics Committee of Nanfang Hospital approved the study and waived patients’ informed consent. We followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines [22].

In the current study, we included hospitalized patients who met all of the following criteria: (1) aged between 18 and 99 years; (2) received at least one prescription of opioid analgesics during the first 30 days of hospitalization; and (3) had at least one SCr test within 7 days prior to opioid analgesic use and at least two SCr tests within any 7-day window after opioid analgesic use. Patients who met one of the following conditions were excluded: (1) end-stage renal disease or receiving maintenance dialysis or received renal transplantation; (2) AKI before or on the day of opioid analgesic use; (3) baseline estimated glomerular filtration rate (eGFR) ≤30 mL/min/1.73 m² or ≥150 mL/min/1.73 m²; (4) use of opioid analgesics on the surgery day only; and (5) use of multiple types of opioid analgesics on the day of the first opioid analgesic prescription or on the day of AKI occurrence. Patients with International Statistical Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes of N17 or O90.4 on admission were considered to have community-acquired AKI and were also excluded. For patients with multiple qualified hospitalizations, only the first hospitalization was included in the analysis.

The primary outcome was the time from opioid analgesic use to the occurrence of HA-AKI. AKI was defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) SCr-based criteria [23]. We used a dynamically screening algorithm to detect AKI, as previously described [2]. In brief, the SCr data were sorted in increasing order of the test time. At any time point t, a reference SCr was defined as the average SCr within 7 days before t (inclusive), and each available SCr data within 7 days after t were compared with the reference. The earliest time that the SCr change fulfilled the Kidney Disease: Improving Global Outcomes (KDIGO) criteria was defined as the onset date of AKI. In this study, urine output criteria were not used because urine volume was not available. The secondary outcome was the time from opioid analgesic use to in-hospital death.

We categorized all prescribed medications according to the Anatomical Therapeutic Chemical (ATC) classification system [24]. We identified all opioid analgesics using the ATC code starting with N02A. The ATC codes of the analyzed opioid analgesics are shown in online suppl. Table S1. Users of opioid analgesics were defined as having at least one prescription of opioid analgesic within the first 30 days during hospitalization. We also calculated the cumulative dose of each opioid analgesic at any given time for each patient.

Baseline comorbidity variables, such as hypertension and diabetes, were defined as presence of the corresponding ICD-10 diagnosis codes (shown in online suppl. Table S2) on or before the use of opioid analgesics, and the comorbidity burden was calculated using the Charlson’s comorbidity index [25]. Patients with an ICD-10-CM diagnostic code of I50.907, I50.103, I50.101, I50.104, or a baseline brain natriuretic peptide of ≥500 ng/L were regarded as having acute heart failure. Baseline procedure variables were defined as presence of the corresponding ICD-9 CM codes on or before the use of opioid analgesics during the index hospitalization. Baseline concomitant medications were defined as presence of the corresponding ATC codes on or before the use of opioid analgesics during the index hospitalization, and common nephrotoxic drugs and their corresponding ATC codes are listed in online supplementary Table S3. The baseline values for laboratory measurements (except for eGFR) were defined as the most recent measurement within 30 days on or before the use of opioid analgesics. Since missing values were common for many laboratory assays and the missing value is usually not random, we transformed each laboratory result into one of the following categories: low, normal, high, and missing (shown in online suppl. Table S4) [26] and used the categorical values for the subsequent analyses.

The baseline characteristics of the study population, stratified by the type of opioid analgesics, were summarized. Continuous variables were presented as the mean ± standard deviation (SD) or median (interquartile range), and categorical variables were presented as numbers (percentages).

Based on previous literature [10, 11, 27‒31], we grouped the opioid analgesics by their indications into the following 2 classes: opioid analgesics for mild-to-moderate pain and those for moderate-to-severe pain. The former included dihydrocodeine, ibuprofen-codeine, and tramadol. The latter included morphine, pethidine, oxycodone, fentanyl transdermal patch, pentazocine, butorphanol, hydromorphone, nalbuphine, and dezocine.

We used the Kaplan-Meier method to estimate the cumulative incidence of HA-AKI after opioid analgesic use. We treated patients as censored at the time of addition of or switch to another opioid analgesic or at the time of the last creatinine measurement, whichever came earlier. We assessed the association between the use of an individual opioid analgesic within each class and the risk of HA-AKI using a Cox proportional hazards model with adjustment for possible confounders selected by the least absolute shrinkage and selection operator (LASSO) regression analyses described below. First, we performed a LASSO logistic analysis to select a set of variables that was associated with the propensity of each specific opioid analgesic and a LASSO survival analysis to select a set of variables that was associated with the risk of HA-AKI. Each LASSO analysis started with a total of 1,035 variables at the baseline, including age, sex, need for surgery, need for intensive care variables for presence of 155 comorbidities, 15 surgical procedures, and 783 concomitant medication, and results from 82 laboratory tests, and a 10-fold cross-validation was used to determine the optimal value of the penalty parameter. Then, we obtained the set of possible confounders by pooling the variables selected from the above LASSO analyses. Last, we estimated the hazard ratio (HR) of each individual opioid analgesic for the risk of HA-AKI using a Cox proportional hazards model with the confounder set as covariates. The opioid analgesics for mild-to-moderate pain and those for moderate-to-severe pain were analyzed separately, as pain severity may confound the analysis, but this information was not available in our study. For both analyses, the most common opioid analgesics (tramadol for mild-to-moderate class and dezocine for moderate-to-severe class) were used as the reference group, and the detailed adjusted variables in the Cox proportional regression models which investigated the risk of HA-AKI in adult opioid analgesic users are presented in online supplementary Table S5. When investigating the risk of HA-AKI and in-hospital death, we also used LASSO cross-validation to select variables. The selected variables in the Cox proportional regression models for the risk of in-hospital death included those associated with the propensity of analgesics or the risk of in-hospital death (see online suppl. Table S6 for the list of variables).

We performed three sensitivity analyses for robustness of the association. First, we restricted the association analysis to the patients with 2 or more days of exposure to opioid analgesic treatment. Second, we assess the association in a 1:1 propensity score-matched cohort selected by the nearest-neighbor matching without replacement and within a specified caliper width (0.2*SD of the logit of the propensity score). For each baseline variable, a standardized mean difference of < 0.10 was considered a satisfactory balance between two groups [32]. Third, we also estimated the association effect using the Fine-Gray model, treating death as a competing risk.

We examined the association between opioid analgesic use and HA-AKI risk among subgroups stratified by age (<60 and ≥60 years), sex, region (South China, Southwest China, East China, and Central and North China), Charlson’s comorbidity index (<6 and ≥6), presence of comorbidities including cancer, chronic kidney disease, acute heart failure, hypertension, diabetes, and liver disease, need for surgery, need for intensive care, concomitant use of other drugs with nephrotoxic effects, and baseline eGFR (<60 and ≥60 mL/min/1.73 m²). An interaction term was added to the Cox model to test for possible effect modification by the grouping factor.

We used a penalized smoothing spline (df = 4) in the Cox models to estimate the dose-response curves of opioid use and the risk of HA-AKI. In the Cox models, the cumulative dose of opioid analgesics was coded as a time-varying variable.

All statistical analyses were performed using R version 3.6.1, with packages “glmnet” and “survival” for the LASSO analyses and survival analyses, respectively. A two-sided p value of <0.05 was considered significant for all tests.

A total of 255,265 in-patients who had used opioid analgesics during hospitalization were included in the analysis (shown in Fig. 1). The mean age was 56.0 (SD 15.5) years, and 54.2% were men. The most common indications of opioid analgesics were for pain management in patients with surgery (116,557 [45.7%]) and cancer (75,353 [29.5%]). The demographic and clinical characteristics of the patients stratified by the types of opioid analgesics used are shown in Table 1. Dezocine (84,943) and tramadol (68,800) were the most frequently prescribed opioid analgesics for treating moderate-to-severe and mild-to-moderate pain, respectively. Therefore, dezocine and tramadol were used as the reference (active comparator) for these two classes of opioid analgesics, respectively, in the subsequent association analyses. Hydromorphone, pentazocine, and butorphanol were most frequently prescribed to patients who had undergone surgery, who constituted 88%, 68.7%, and 68.5%, respectively, of the total users of the corresponding opioid analgesics. Cancer was prevalent among the users of fentanyl transdermal patch (75.5%), butorphanol (44.5%), and dihydrocodeine (39.0%) but not among morphine users (16.8%).

A total of 11,601 HA-AKI events were detected in the study population. Among the users of opioid analgesics for moderate-to-severe pain, morphine users had the highest cumulative incidence of HA-AKI (shown in online suppl. Fig. S1), with a cumulative incidence of 41.3% (95% confidence interval [CI] 36.7%–45.7%) on the 30th day of hospitalization compared with 16.6% (95% CI: 15.2%–18.1%) in dezocine users. Compared with dezocine, exposure to morphine was associated with a 56% higher risk of HA-AKI (HR 1.56, 95% CI: 1.40–1.78) after adjusting for the confounders (shown in Table 2). Meanwhile, there was no significant difference in the risk of HA-AKI between the users of dezocine and other opioid analgesics of the same class, including pethidine, oxycodone, butorphanol, pentazocine, fentanyl transdermal patch, hydromorphone, and nalbuphine. We did not observe any significant difference in the risk of HA-AKI among the users of the opioid analgesics for treating mild-to-moderate pain (shown in Table 2).

In the first sensitivity analysis that restricted to patients with at least 2 days of exposure to opioid analgesics, exposure to morphine remained to be associated with a significantly increased risk of HA-AKI compared with dezocine (HR 1.60, 95% CI: 1.40–1.84) (shown in online suppl. Table S7). Similar to the main analysis, there was no significant difference in the risk of HA-AKI among the users of other types of opioid analgesics in patients with moderate-to-severe or mild-to-moderate pain. In 5,856 propensity score-matched pairs of morphine and dezocine users, the baseline characteristics were well balanced, with a maximal value of standardized differences of 0.049 among all the variables considered (shown in online suppl. Table S8). In the other propensity score-matched pairs of opioid analgesic users, the values of standardized differences among the variables considered were all less than 0.1 (shown in online suppl. Table S9–S17). In the propensity-matched analyses, morphine (HR 1.80, 95% CI: 1.48–2.20) but not other opioid analgesics was associated with a significantly increased risk of HA-AKI compared with dezocine (shown in Table 3). There was no significant difference in the risk of HA-AKI among the users of opioid analgesics for treating mild-to-moderate pain. In the competing-risk analysis, compared with dezocine, a subdistribution HR of 1.41 for HA-AKI (95% CI: 1.28–1.54) was observed for morphine use, and there was no significant difference in the risk of HA-AKI among the users of opioid analgesics for treating mild-to-moderate pain (shown in online suppl. Table S18).

A similar association between morphine use and the risk of HA-AKI was observed among different subgroups stratified by age, sex, region, Charlson’s comorbidity index, baseline eGFR, presence of diabetes, chronic kidney disease, hypertension, acute heart failure, cancer, and liver disease, surgery, need for intensive care, concomitant use of angiotensin-converting enzyme inhibitors or angiotensin receptor blockers, non-steroidal anti-inflammatory drugs, and contrast media (shown in Table 4). All p values for interaction were larger than 0.05.

The distribution of the cumulative dose of the opioid analgesics during the first 30 days of hospitalization is presented in online supplementary Figure S2. There was a significant nonlinear dose-response relationship between morphine use and the risk of HA-AKI (shown in Fig. 2). The risk of HA-AKI was increased even at a low cumulative dose of morphine and plateaued around 55 mg.

A total of 697 in-hospital deaths occurred among 16,375 morphine users. The incidence of in-hospital deaths was 18.8% and 2.4%, respectively, in morphine users with and without HA-AKI. Among morphine users, HA-AKI was associated with a significantly increased risk of in-hospital death (HR 3.54, 95% CI: 2.89–4.33) after adjustment for age, sex, hospital, division, need for intensive care, surgery, Charlson’s comorbidity index, and other variables (shown in online suppl. Table S19).

This large multicenter real-world study evaluates, for the first time, the risk of HA-AKI among adult users of different opioid analgesics. Among adult users of opioid analgesics with moderate-to-severe pain, morphine users had the highest risk of HA-AKI, which was 56% higher than that in users of dezocine, a most commonly used opioid analgesic in our study population. The risk of HA-AKI associated with morphine use was dose-dependent. Furthermore, the risk of HA-AKI did not differ significantly among the users of other types of opioid analgesics in patients with moderate-to-severe or mild-to-moderate pain. The results remained consistent across the sensitivity analyses.

There are at least a dozen opioid analgesics on the market, and currently, there is a lack of evidence to guide the physicians on the choice of opioid analgesics in patients who are at high risk of AKI. Previous studies on the use of opioid analgesics and AKI mainly focused on morphine and found that the use of morphine was associated with an increased risk of AKI compared with non-opioid users in elders (mean age, 82 years) [14] or patients with acute heart failure [15]. The effects of opioid analgesics other than morphine on the risk of AKI have not been examined previously, except for a few case reports [16‒20]. In the current study, we have compared the effects of all types of opioid analgesics commonly used in Chinese hospitals on the risk of HA-AKI among all hospitalized patients, using the most common opioid analgesics as the reference. Consistent with the previous reports, our study showed that morphine use was associated with an increased risk of HA-AKI (HR 1.56, 95% CI: 1.40–1.78) in patients with moderate-to-severe pain after adjusting for the confounders, and HA-AKI significantly increased the risk of in-hospital deaths among morphine users. Meanwhile, the risk of HA-AKI was not significantly different among the users of non-morphine opioid analgesics in patients with moderate-to-severe pain. In particular, there were nominal differences in the risk of HA-AKI among the users of dezocine, pethidine, and oxycodone, the three most commonly used opioid analgesics for treating moderate-to-severe pain, as the CIs of the estimated HRs were very small.

Patients need opioid analgesics and those who do not are intrinsically different in clinical features and AKI susceptibilities. To minimize the selection bias and the effect of possible confounders in our analysis, we performed analyses separately for opioid analgesics used for treating mild-to-moderate pain and moderate-to-severe pain and vigorously adjusted for all possible confounders in the analyses. We used the LASSO shrinkage method to select >300 demographical and clinical variables that may be associated with the propensity of analgesic use or the risk of AKI and adjusted these variables in our subsequent association analyses. We also performed sensitivity analyses using the propensity score-matched cohorts that have a balanced distribution of the clinical variables among users of different opioid analgesics. Similar association effects of opioid analgesics on the risk of HA-AKI were observed among different subgroups and among different sensitivity analyses, suggesting the robustness of our findings.

This study has several strengths, including the large-scale and multicenter design, and the authors were able to calculate relatively precise estimates for the risk of AKI among adult opioid analgesic users. The availability of rich clinical data enabled the control of a number of confounders that might bias the association between the risk of HA-AKI and opioid analgesic exposure. The active control design and use of the propensity score matching cohort mitigated the risk of indication bias [33].

Our study has several limitations. First, our data lacked pain scores [34] for the patients, which may confound our analyses. However, we have performed our analyses separately for opioid analgesics used for treating mild-to-moderate and moderate-to-severe pain. Second, we only assessed the short-term effects of opioid analgesics on the risk of HA-AKI during hospitalization. The effect of long-term use could not be evaluated due to the lack of out-patient prescription data. Third, the study cohort only included Chinese adults. Whether the results could be generalized to other ethnic populations requires further validation. Fourth, at least 14% of the study participants were patients with cancer or trauma. SCr levels in some of these patients may not necessarily reflect their true kidney function. Fifth, the detection of HA-AKI relies on multiple measurements of SCr. Exclusion of the patients without multiple SCr measurements may lead to an overestimation of the overall AKI incidence. Last, as in any observational study, confounding by unknown factors cannot be fully excluded.

The use of morphine was associated with an increased risk of HA-AKI in adult patients with moderate-to-severe pain, while the risk of HA-AKI did not differ significantly among the users of other types of opioid analgesics in patients with moderate-to-severe or mild-to-moderate pain. Opioid analgesics other than morphine should be chosen preferentially in adult patients with high risk of HA-AKI when treating moderate-to-severe pain. The findings should be interpreted cautiously, given the potential for residual confounding.

Dr. Gang Xu was not available to confirm co-authorship, but the corresponding author, Dr. Fan Fan Hou, affirms thatDr. Gang Xu contributed to the paper, had the opportunity to review the final version to be published, and guarantees Dr. Gang Xu co-authorship status and the accuracy of the author contribution and conflict of interest statements.

The study follows the principles of the Helsinki Declaration. The study protocol was approved by the Medical Ethics Committee of Nanfang Hospital, Southern Medical University (approval number: NFEC-2019-213), which waived the requirement for patients’ informed consent. This study was also approved by the China Office of Human Genetic Resources for Data Preservation Application (approval number: 2021-BC0037). The Medical Ethics Committee of Nanfang Hospital specifically approved the informed consent waiver because this study extracted anonymous data from electronic health databases.

The authors have no conflicts of interest to declare.

This work was supported by the National Natural Science Foundation of China (81970586 to X.X.), Key Technologies R&D Program of Guangdong Province (2023B1111030004 to F.F.H.), the National Natural Science Foundation of China (Key Program) (82030022 to F.F.H.); the Program of Introducing Talents of Discipline to Universities, 111 Plan (D18005 to F.F.H.); Guangdong Provincial Clinical Research Center for Kidney Disease (2020B1111170013 to F.F.H.).

F.F.H. and X.X. had full access to all the data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis. F.F.H., X.X., and M.P. conceived and designed the study. F.F.H. took the lead in drafting the manuscript and received major funding for the study. X.X. and M.P. provided substantial scientific input in statistical methods and data analysis. F.F.H., Y.Z., Y.S., H.L., J.Z., Y.K., G.L., Y.H., H.L., Q.W., C.C., B.L., Q.Y., G.S., Y.Z., J.W., G.X., H.X., Y.T., and M.G. contributed to the data collection. S.N., Y.L., Y.C., P.G., L.S., and Y.L. prepared and cleaned the data. All authors contributed to the interpretation of data, provided critical revisions to the manuscript, and approved the final draft.