Research design

This study was a retrospective cohort study (data science) that applied medical big data (TheBD: The Tokyo University Health Economy Big Data) for real-world evaluation. The study data were collected from April 2012 to December 2020, and the target population included patients diagnosed with CKD (International Classification of Diseases, 10th edition [ICD-10] code: N18) who visited medical institutions (examination and treatment intervention) in Japan. CKD is defined as ‘renal impairment evident by urinalysis, blood tests, or imaging’; ‘glomerular filtration rate (GFR) less than 60 ml/min/1.73 m2’, or both of these findings for at least 3 months. The exclusion criteria were as follows: (1) individuals aged < 20 years who had congenital renal disease; (2) individuals receiving renal replacement therapy; and (3) individuals with cancer (within 3 years prior to observation), except for renal cancer, which was excluded from the sample of patients with cancer. Samples for which the use of iodine contrast media could not be correctly determined were also excluded. We adhered to the STROBE guidelines for this cohort study.

The cohort was divided into contrast agent therapy (CAT, contrast-enhanced treatment) and non-contrast agent therapy (non-CAT, non-contrast-enhanced treatment) groups for laboratory and therapeutic interventions while considering the risk factors for CKD, such as hypertension, diabetes, medications (e.g., analgesics and antihypertensive drugs), age, and sex. Renal function decline was calculated from the correspondence table between the CKD stage (KDIGO 2012: G1–G5) and GFR reference values (CKD Practice Guide 2012: mL/min/1.73 m2) [14, 15]. The endpoints were the number of contrast exposures, regardless of contrast route and dose, and renal function derangement. They were analysed in units of annual averages divided by the observation period, accounting for the timing of reciprocal occurrences. The framework was a long-term longitudinal study based on disease characteristics and study objectives. The average analysis period was 5 years, with a minimum observation period of 6 months or longer.

Study data source

This study used a large database that includes medical service data examined by a specialised public organisation (Social Insurance Medical Fee Payment Fund), in accordance with the format stipulated by the Ministry of Health, Labor and Welfare (MHLW Notification: Vol. 0831 No. 1). We selected medical economic big data (TheBD; S1 Table), [16, 17] which included medical service bills in Japan between April 2012 and March 2021, representing the coverage for 7 million insured patients.

The present study was conducted in accordance with the principles embodied in the Declaration of Helsinki. This study received comprehensive approval in March 2019 from the Institutional Review Board of the University of Tokyo Hospital (screening no: 2018167NI, approval date: March 26, 2019). Because we used database records (anonymized) for analysis, the requirement for the acquisition of informed consent from patients was waived at the above Institutional Review Board (opt-out format).

The largest proportion of the sample was from the year 2016 (22.1%). Medical information accounted for 6.18 million results while dispensing information accounted for 6.20 million results (including duplications). The patient-based hospitalisation rate was 13.5% (including duplications), and the average percentage of male patients for all years was 46.8%. This database is updated every 6 months. All data on disease name, testing, medication, surgery, and any other medical interventions with dates of initiation and related costs are linked in chronological order using unique IDs for each patient. During each biannual update, the transfer of data for insured persons is managed, and adjustments are made according to the allocation of medical facilities. TheBD has been used in several studies evaluating the economic aspects of medical interventions (S2 Table).

Selection conditions for research subjects

For cases that met the abovementioned selection criteria but did not meet the exclusion criteria, we defined the observation period to be at least 6 months from the date of the first diagnosis of CKD stage. Injuries were identified by the ICD-10 code (N18) and CKD stage (injury name).

The analysis period was from the date when the CKD stage injury was first identified (index day) or the date of data generation, whichever was earlier, to the date when renal replacement therapy was initiated. We observed hemodialysis, peritoneal dialysis, and renal transplantation interventions as renal replacement therapies. For patients who did not receive renal replacement therapy, the analysis period was up to the date of the last identifiable CKD stage injury (end date).

Data extraction was performed on the six components (conditions) listed in the ‘Extraction conditions’ table shown in Fig. 1. We first selected patients (15095 cases) with renal failure that met the selection criteria from the medical big data (Fig. 1). We then excluded those that met the exclusion criteria, for a final sample of 1366 cases. We classified this population into CAT and non-CAT groups according to contrast exposure. The patient background in both groups was adjusted for major risk factors for renal function by applying the propensity score matching (PSM) method.

Fig. 1

Structure of this study: basic flow of data set and main analysis conditions

PSM, propensity score matching

Thus, this study included 333 patients in the CAT and non-CAT groups, totalling to 666 patients. In this study, most CATs were imaging tests. We also performed a stratified analysis focusing on the group with altered renal function (260 patients; 39.0%), while taking the acceleration of CKD progression into account. For the stratified analysis (260 patients), the decline in renal function group was selected from the basic analysis cohort (666 patients). After processing PSM, the analysis included 63 patients in each group, totalling 126 patients.

We calculated the cut-off value of renal function decline for CAT from the receiver operating characteristic (ROC) curve in the stratification analysis.

Evaluable interventions and precautionary conditions

The interventions evaluated in this study were practices involving iodine contrast media. All iodine contrast media approved in Japan were included in the evaluation.

Iodine contrast media are classified into two types: water-soluble and oil-based. Water-soluble contrast media are used in patients with normal renal function. Approximately 90% is excreted as urine from the kidneys at approximately 6 h after injection. In contrast, oil-based iodine contrast media can remain in the body for a long period. These two types of contrast media are used separately according to the purpose of the examination. Water-soluble contrast media are selected in most cases. Water-soluble contrast media can be further classified into ionic and non-ionic contrast media. Compared with ionic contrast media, non-ionic contrast media have lower osmotic pressure and are more similar to blood, which alleviates the symptoms of adverse reactions, such as the burning sensation experienced during injection. In Japan, non-ionic contrast media are most commonly used in angiography.

During the examination of contrast exposure and its impact on renal function, the amount of contrast media injected into the body is of importance. However, the data source for this study is based on information from medical claims, which provides information on product consumption but not on internal infusion volume. In addition, information on weight and detailed imaging type of the subject cases is essential for determining the internal infusion volume. However, due to the large uncertainty, there were limitations regarding the organisation of this information.

The rehydration (saline) volume was categorised by drug category (code 331), and the cumulative volume (in mL/year) within the observation period was calculated. Consistency was also maintained for rehydration on a yearly basis in relation to the analysis of time-series changes in renal function. This volume was calculated from the volume set for each drug name, as there was variation in the volume or number of times that it was used, depending on the case or prescription. Furthermore, if the frequency was two or more times, it was further corrected by considering the frequency in addition to the calculated amount used. The figures for this calculation are the product consumption, not the injectable volume.

Main analysis methods and assessment of renal function

The Mann–Whitney U test was used to evaluate the effect of iodine contrast media use on renal function, and the chi-square test was used along with the previously described test to compare patient background. The significance level was set at 5%. We calculated Pearson’s correlation coefficient to assess the correlation between the number of contrast media used and the decline in renal function. Multivariate analysis of risk factors for renal function was performed using logistic regression analysis. Statistical analysis was performed using SPSS version 26.0 (IBM Corp., Armonk, New York). The figures in the tables are presented as means and standard deviations, whereas the figures are presented as standard errors.

PSM was used to minimise bias related to patient background. Therefore, the forced entry method was chosen for the multivariate logistic regression model. The independent variables (covariates) were age, sex, hypertension/diabetes, use of NSAIDs, antimicrobial agents, antineoplastic agents, and others. Propensity scores for each case were calculated, and a 1:1 matching method was applied to align the sample numbers and confirm the data distribution and balance in both groups. Among the independent variables (covariates), age was defined as the date of CKD stage; hypertension was diagnosed within the observation period; and the use of NSAIDs, antimicrobial agents, and antineoplastics within the observation period was defined as the scope of analysis.

The GFR in each case was converted from the GFR category correspondence table of CKD severity classification based on the observed CKD stage in ICD-10 codes in this study. Furthermore, we estimated the decline in renal function associated with changes in CKD stage from the CKD stages and GFR levels in previous studies [18,19,20,21]. The underlying data used for the conversion are shown in Table 1. The annual change in estimated GFR was calculated as follows: [ΔGFR: estimated GFR (index day) - estimated GFR (end date)] / [duration: observation period]. When some information regarding the CKD stage was missing, estimates were made based on pre- and post-stage data.

Table 1 CKD stage to GFR conversion