ABSTRACT
INTRODUCTION
Epidemiological studies using large-scale administrative databases have been attracting worldwide attention in recent years [1]. Administrative databases enable researchers to conduct large-scale studies in real-world settings [2]. However, such databases are used under the assumption that they convey accurate information on health conditions and service provision [3]. Validating the data stored in administrative databases is crucial because the misclassification of exposures or outcomes may lead to biased results when performing database studies [4]. Previous database studies using diagnostic codes for hyponatremia have reported the validity of code-based diagnosis [5, 6].
Several validation studies on hyponatremia have been conducted in different countries [7-10]. A study on the International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes for hyponatremia in an outpatient setting revealed a sensitivity of 3.5% and specificity of >99% for identifying hyponatremia (serum sodium <136 mmol/l) [7]. A study on hyponatremia in inpatients reported a sensitivity of 1.7% and a specificity of >99% [8]. The International Classification of Diseases, Tenth Revision (ICD-10) codes for hyponatremia have been reported to have a similarly low sensitivity and high specificity [9, 10].
Bạn đang xem: Diagnostic validity and solute-corrected prevalence for hyponatremia and hypernatremia among 1 813 356 admissions
Evidence is lacking on validation studies of the ICD-10 codes for hypernatremia, another important electrolyte abnormality. Furthermore, most current laboratory measurement systems use indirect ion-specific electrodes [11], with which the presence of high levels of solutes such as blood glucose, triglycerides, and total protein [12-14] can disrupt the correlation between sodium concentration and plasma tonicity, leading to inaccurate assessments of osmolality [11, 15]. In particular, severe hypertriglyceridemia or paraproteinemia can cause measurement artifacts in the dilution process of indirect ion-specific electrodes, which may lead to an underestimation or overestimation of tonicity-based sodium disorders [16]. Although a few articles reported solute-corrected prevalence of hyponatremia [17, 18], possible effects of the corrections on the prevalence of hyponatremia and hypernatremia have not been discussed in general hospitalizations.
This study examined the accuracy of diagnoses of hyponatremia and hypernatremia in hospitalizations using a hospital-based database. We also evaluated the possible effects of these corrections on the prevalence of hyponatremia and hypernatremia.
MATERIALS AND METHODS
Data source
We used data from the JMDC hospital-based database (JMDC Inc., Tokyo, Japan). The details of this database have been described previously [19]. Briefly, the database contains the Japanese Diagnosis Procedure Combination (DPC) data, claims data, and laboratory values obtained from ∼95 hospitals. The Japanese government introduced the DPC system to standardize the electronic claims system and to realize transparency of hospital performance [20]. The DPC data of the JMDC database are recorded in a manner similar to that stored in other databases and include the following items: patient demographics, detailed clinical information on diseases, patient statuses at admission and discharge, diagnoses, procedures, and medications. Diagnoses are recorded based on ICD-10 codes and Japanese free text, recorded by the attending physicians [21]. Suspected diagnoses are recorded and are denoted accordingly. Six categories of diagnoses exist: “main diagnosis,” “admission-precipitating diagnosis,” “comorbidities at admission,” “complications occurring after admission,” “most resource-consuming diagnosis,” and “second most resource-consuming diagnosis” [20]. The DPC data have a distinctive property in that the main diagnosis, comorbidities at admission, and complications during hospitalization are clearly distinguishable among the recorded diagnoses [20, 21]. An increasing number of validation studies on procedure codes and disease names in the DPC data have been published [22-26].
Study population
Using the JMDC database, we identified patients aged ≥18 years who were discharged between 1 April 2014 and 31 August 2023 and had at least one measurement for serum sodium concentration during hospitalization. The exclusion criteria were as follows: absence of information on age, sex, whether the admission was unscheduled, ambulance use, consciousness level on admission, or admission-precipitating diagnosis and maximum sodium concentration of >230 mmol/l during hospitalization, and minimum sodium concentration of <90 mmol/l during hospitalization. The criteria for erroneous values were based on case reports of hypernatremia (209 mmol/l) [27] and hyponatremia (98 mmol/l) [28], with an additional 10% margins. Repeated hospitalizations for a single patient were included and analyzed independently.
Study outcomes and variables
The following patient information was extracted from the database: age; sex; body mass index (BMI); main diagnosis, admission-precipitating diagnosis, comorbidities at admission, and complications occurring after admission; smoking status (current/past or non-smoker); consciousness on admission according to the Japan Coma Scale; activities of daily living according to the Barthel index [29], and in-hospital death.
We extracted all serum sodium concentration test results during hospitalization and summarized them as daily data. We primarily defined hyponatremia as a serum sodium concentration of <135 mmol/l [10, 30-32] and also performed analyses with <130 and <125 mmol/l for moderate and severe hyponatremia, respectively [32]. Hypernatremia was defined as a serum sodium concentration of >145 mmol/l [33-35], as well as >150 and >155 mmol/l for moderate and severe hypernatremia, respectively. In cases where a patient underwent multiple examinations on a single day, the most extreme measurements were recorded.
The ICD-10 codes of E87.1 and E87.2 were used to identify hyponatremia and hypernatremia from the DPC data, respectively. We categorized each hospitalization based on the ICD-10 code of the admission-precipitating diagnosis as follows: infectious (A00-B99), neoplasms (C00-D48), hematological (D50-D89), endocrinological (E00-E90), mental (F00-F99), neurological (G00-G99), ophthalmological (H00-H59), otological (H60-H95), cardiovascular (I00-I99), respiratory (J00-J99), digestive (K00-K93), dermatological (L00-L99), musculoskeletal (M00-M99), genitourinary (N00-N99), pregnancy (O00-O99), perinatal (P00-P96), congenital (Q00-Q99), symptoms/signs (R00-R99), injury/poisoning (S00-T98), new diseases (U00-U99), external causes (V00-Y98), and examinations (Z00-Z99).
The Institutional Review Board of the Graduate School of Medicine of The University of Tokyo (2018030NI) approved the study protocol. Owing to the anonymous nature of the data, the requirement for informed consent was waived.
Statistical analysis
We first summarized the background characteristics of the eligible population. We subsequently categorized individuals into two groups based on the occurrence of hyponatremia (<135 mmol/l) during hospitalization. We compared the distributions between those with and without hyponatremia using the Wilcoxon rank-sum test for continuous variables and χ2 tests for categorical variables. Similarly, we classified the eligible individuals into two groups based on the occurrence of hypernatremia (>145 mmol/l) during hospitalization and compared their characteristics. We also identified the 10 most frequent admission-precipitating diagnoses among all the patients, those with hyponatremia and those with hypernatremia, based on the first three digits of the ICD-10 codes.
Using laboratory data as the reference standard, we examined the validity of DPC data for identifying hyponatremia in three different scenarios corresponding to the timing of diagnosis: using all six categories of diagnoses to identify hyponatremia observed at least once during hospitalization, using diagnoses present at the time of admission (i.e. “admission-precipitating diagnosis” or “comorbidities at admission”) to identify hyponatremia observed on the day of admission, and using diagnoses occurring after admission (i.e. “complications occurring after admission”) to identify hyponatremia observed at least once on the second day of hospitalization or later. For the last two scenarios, those with no measured sodium levels at the respective time points were excluded from the analyses. We calculated the sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR−), diagnostic odds ratio (DOR), positive predictive value (PPV), and negative predictive value (NPV) of DPC data-based identification against the reference thresholds of <135, <130, and <125 mmol/l serum sodium. These procedures were repeated for hypernatremia, with the reference thresholds of >145, >150, and >155 mmol/l. An overview of the validity indices in this study is shown in Fig. 1. We also performed analyses stratified by age (the median of eligible individuals) and the admission-precipitating disease (neoplasms, cardiovascular, respiratory, and others). As a supplementary analysis, we used 120, 115, 110, and 105 mmol/l as cutoffs for hyponatremia. We provided the statistics for these cutoffs because sodium levels of ≥120 mmol/l are considered a safe range for developing osmotic demyelination syndrome [36] , while sodium levels of <105 mmol/l suggest the highest-risk group for developing osmotic demyelination syndrome [37]. We performed two sensitivity analyses: one limiting hospitalizations to the first one during the research period for each patient, and the other excluding individuals with diabetes (ICD codes E10-E14), hypertriglyceridemia (E78.1), or multiple myeloma (C90.0).
Considering that blood glucose, triglyceride, and total protein levels reportedly affect sodium concentrations [12-14], we calculated statistics before and after correction for these values among individuals with data for all these parameters. Marked hyperglycemia [11, 15], hypertriglyceridemia or paraproteinemia [16], and hypoproteinemia [17] can disrupt the correlation between sodium concentration and plasma tonicity. We used laboratory values for these three items, which were measured on the same day as the sodium measurement of interest. In cases where a patient underwent multiple examinations on a single day, the most extreme values for glucose, triglycerides, and total protein were recorded. We corrected sodium concentrations as follows [12-15]:
where [Na+] indicates measured sodium concentration, glucose refers to blood glucose concentration (mg/dl), TG refers to triglyceride concentration (g/l), and TP stands for total protein level (g/dl) (equations for different units are shown in the Supplementary Methods).
The linear function was used for glucose-corrected and total-protein-corrected sodium concentrations across all ranges of each value, while the cubic function was applied for triglyceride-corrected sodium concentrations exceeding 1500 mg/dl, according to previously described methods [12-15].
Because of the large sample sizes, we did not perform statistical testing. Stata version 18 (StataCorp, College Station, TX, USA) was used to conduct all statistical analyses.
RESULTS
Study population
We extracted data from 1 825 873 hospitalizations with recorded serum sodium levels in the JMDC database. After excluding 12 517, we obtained 1 813 356 records from 1 156 291 patients for analysis (Fig. 2). We identified 419 470 (23.1%) and 132 563 (7.3%) hospitalizations with hyponatremia (serum sodium level of <135 mmol/l) and hypernatremia (serum sodium level of >145 mmol/l), respectively. Conversely, 18 378 (1.0%) and 2950 (0.2%) had ICD-10-based hyponatremia and ICD-10-based hypernatremia, respectively. Patients with hyponatremia (<135 mmol/l) were more likely to be older, leaner, exhibit disturbed consciousness, and die during hospitalization than those without hyponatremia (Table 1). These characteristics were also observed in patients with hypernatremia based on sodium levels (Table 2). However, while the proportion of males was higher in the group with hyponatremia (<135 mmol/l) compared with the group without, the hypernatremia group (>145 mmol/l) had a higher proportion of females than the non-hypernatremia group. The most frequent admission-precipitating diagnosis among all hospitalizations was heart failure (ICD-10, I50), followed by cerebral infarction (I63) and angina pectoris (I20), as shown in Table 3. Among patients with hyponatremia (<135 mmol/l) and hypernatremia (>145 mmol/l), the most frequent admission-precipitating diagnosis was heart failure (ICD-10, I50), followed by pneumonitis due to solids and liquids (J69).
Validity of Diagnosis Procedure Combination data for identifying a laboratory diagnosis of hyponatremia
Table 4 shows the validity indices of the DPC data for the identification of hyponatremia. Overall, the ICD-10 code (E87.1) showed a low sensitivity and high specificity; the sensitivity was 4.1%, and the specificity was 99.9% to identify serum sodium levels <135 mmol/l during hospitalization. The LR+, LR−, and DOR were 41.1, 0.96, and 42.8, respectively. The PPV was 92.5%, and NPV was 77.6% in this research context. When moderate and severe hyponatremia were defined by serum sodium levels of <130 and <125 mmol/l, the sensitivities increased to 11.4% and 23.8%, respectively, while maintaining specificity >99% across the various cutoff points. Compared to using all diagnoses to identify hyponatremia during hospitalization, when using ICD-10-based diagnosis, the sensitivity was higher on the day of admission and lower post-admission day.
Table 5 presents the results of stratified, supplementary, and sensitivity analyses. In comparison to the analysis of the patients aged <75 years, the analysis of those aged ≥75 years yielded a higher sensitivity, while the specificity was similarly high. The sensitivity was lowest for patients with neoplasms, followed by those with cardiovascular and respiratory diseases. When the cutoffs for hyponatremia were lowered to 120, 115, 110, and 105 mmol/l, the sensitivity increased, and the specificity decreased as the cutoff became more extreme, similar to the observations in the main analysis. The results of the sensitivity analyses were similar to those of the primary analysis.
Validity of Diagnosis Procedure Combination data for identifying a laboratory diagnosis of hypernatremia
Table 6 presents the validity of the DPC data for hypernatremia. Overall, the ICD-10 code (E87.2) exhibited low sensitivity and high specificity. Identifying serum sodium levels of >145 mmol/l during hospitalization yielded a sensitivity of 2.2% and specificity of >99.9%; the LR+, LR−, and DOR were 350, 0.98, and 358, respectively; and the PPV and NPV in this research context were reported at 96.5% and 92.8%, respectively. When serum sodium levels >150 and >155 mmol/l were used to define moderate and severe hypernatremia, the sensitivities increased to 5.9% and 9.3%, respectively, while maintaining specificity at 99.9%. Compared with using all diagnoses during hospitalization to identify hypernatremia, the sensitivity was higher on the admission day and lower post-admission when using an ICD-10-based diagnosis.
The results of stratified and sensitivity analyses are shown in Table 7. Compared with the analysis of younger patients, the analysis of older ones resulted in a generally higher sensitivity, while the specificity remained similarly high. Sensitivity was lowest for patients with neoplasms, followed by those with cardiovascular and respiratory diseases. The results of the sensitivity analyses were similar to those of the primary analysis.
Identification of hyponatremia and hypernatremia when sodium levels are corrected for glucose, triglyceride, and total protein levels
Figure 3 and Tables 8 and 9 present results with and without corrections for glucose, triglyceride, and total protein levels among individuals with available data for each parameter. Both when sodium levels were lowest and highest, mean sodium levels became higher with corrections for glucose or triglycerides, while the corresponding values became lower with corrections for total protein (Table 8). Although the sensitivity and specificity remained largely unchanged after the corrections (Table 10), the prevalence of hyponatremia or hypernatremia was affected by the corrections, and the extent of this effect varied among the types of correction.
Glucose correction decreased the prevalence of hyponatremia, while total protein correction increased that of hyponatremia (Fig. 3). Among individuals with data for glucose (N = 1 146 779), triglycerides (N = 450 019), and total protein (N = 1 486 592) when the sodium levels were lowest during hospitalization, 41 546 (15.2%) of 273 340, 184 (0.2%) of 88 031, 111 (0.03%) of 342 726 patients with hyponatremia without each correction were diagnosed as not having hyponatremia with correction (corrected sodium ≥135 mmol/l), respectively. The opposite occurred with these corrections. Among the individuals with available values for glucose and total protein when the sodium levels were lowest during hospitalization, 19 612 patients (2.2%) of 873 439 and 330 729 patients (28.9%) of 1 143 866 without hyponatremia without each correction were diagnosed as having hyponatremia with correction (corrected sodium <135 mmol/l), respectively. Overall, while the correction for glucose or triglyceride levels in the prevalence of hyponatremia resulted in a net decrease (from 23.8% to 21.9% and from 19.562% to 19.521%, respectively) that for total protein levels resulted in a net increase (from 23.1% to 45.3%). Conversely, while the correction for glucose or triglyceride levels, when the sodium levels were highest during hospitalization, led to a net increase in the prevalence of hypernatremia (from 7.1% to 10.9% and from 4.961% to 4.963%, respectively) that for total protein levels led to a net decrease (from 6.9% to 3.6%). The prevalence of hyponatremia or hypernatremia was not markedly affected after performing sodium correction for high triglycerides in 361 and 71 individuals in the analyses of hyponatremia and hypernatremia, respectively.
DISCUSSION
Xem thêm : Japanese Garlic Rice
In the present observational study using real-world data with laboratory values, we revealed that the ICD-10-based hypernatremia had a low sensitivity and high specificity to identify hypernatremia defined by serum sodium levels of >145 mmol/l. To the best of our knowledge, this is the first study to present such results using nearly 1 million admissions. In addition, we confirmed that corrections by blood glucose, triglycerides, and total protein affected the prevalence of hyponatremia and hypernatremia. By contrast, they did not largely affect the validity of both sodium abnormalities.
The validity of the ICD-10 codes for hyponatremia has been reported in several previous articles, and our analytical results agree with their findings [9, 38]. Lowering the threshold sodium value to include cases with only severe hyponatremia resulted in an increased sensitivity with a minimal impact on specificity. This finding was also consistent with those of previous studies [9, 38]. We also analyzed hyponatremia cutoffs at 110 and 105 mmol/l, ranges not covered in previous studies but important because they pose an increased risk for osmotic demyelination syndrome [37]. Furthermore, our finding of a higher sensitivity of the ICD-10 code among older patients was also observed in one previous study [10]. Notably, the present study found that the ICD-10-based diagnosis after admission had a lower sensitivity than the diagnosis present at admission. This trend was consistent with what was reported in another study [24]. The low sensitivity of the hyponatremia codes suggests that the ICD-10-based hyponatremia diagnosis should not be used for the calculation of prevalence, incidence, and absolute risk difference. While ruling in patients using ICD-10-based hyponatremia may be possible due to high LR+, ruling out patients using the same metric would be virtually impossible as the LR− value was almost one [39]. However, the high specificity and PPV indicate that the calculation of a relative risk could be acceptable as long as the disease is recorded in different groups at the same rate [40, 41]. As an example for the underestimation of hyponatremia prevalence, a previous study using the ICD-10 code reported a lower prevalence of hyponatremia (2%-6%) [5] compared to our study (23%) and another study (14%) [10], both of which used real sodium values. This study may have underreported the incidence of hyponatremia due to the use of the ICD-10 code [5]. However, the seasonality of hyponatremia, described in terms of odds ratios [5], remains interpretable because these ratios are relative indices instead of absolute ones.
Our analysis of the ICD-10 codes for hypernatremia in this study showed that, similar to hyponatremia, the registered disease code had a low sensitivity and a high specificity. We also observed a higher sensitivity for identifying more severe hypernatremia. This suggests that clinicians may have registered hypernatremia only when the sodium levels are particularly extreme. Similar to hyponatremia, the ICD-10-based diagnostic record of hypernatremia in the DPC data should not be used to calculate prevalence, incidence, or absolute risk difference due to its low sensitivity. Similar to ICD-10-based hyponatremia diagnosis, ruling in patients using ICD-10-based hypernatremia may be possible, whereas ruling out patients using the same metric would be impossible. However, the high specificity and PPV indicate that it may be acceptable to use the recorded diagnosis to calculate a relative risk in circumstances where the disease names are registered with the same probability in different patient groups. The validity of hypernatremia recorded in administrative data has not been evaluated in previous literature. Thus, further studies are warranted to confirm our findings in other countries and databases.
Evidence on hypernatremia is limited. Before databases with sodium levels were available, most articles were case reports or case series of extreme hypernatremia [27, 42, 43]. A cohort study published nearly two decades ago showed that hypernatremia was associated with older age, disturbed consciousness, lower BMI, and in-hospital mortality [44]. Recent studies utilizing data on sodium levels from 2 million patients have advanced our understanding, showing that hypernatremia is a significant determinant of hospital disposition and highlighting that extremely high sodium values predict in-hospital mortality [45, 46]. Our study contributes to this growing body of literature by demonstrating the low sensitivity of the ICD-10 diagnostic code for hypernatremia, underscoring the need for improved coding practices and the potential for enhanced patient outcomes through better identification and management of hypernatremia. Our study also uncovered that hypernatremia, like hyponatremia, was associated with increased in-hospital mortality and unscheduled hospitalization. Specifically, the coexistence of hyponatremia was associated with a 3.5-fold risk of in-hospital death, while hypernatremia was associated with a 5.3-fold risk of in-hospital death, indicating that the association for mortality may be stronger with hypernatremia than with hyponatremia. In the future, databases with available electrolyte levels, as in these studies and ours, will reveal other aspects of clinical importance of hypernatremia.
While corrections for glucose, triglyceride, and total protein levels affected the prevalence of hyponatremia and hypernatremia, they did not greatly affect the diagnostic ability of ICD-10 codes. Because the correction for glucose and triglycerides resulted in a mean increase in sodium levels and that for total protein resulted in a mean decrease in sodium levels (Table 8) regardless of highest or lowest sodium levels, correction for glucose and triglycerides led to a decrease in hyponatremia prevalence and an increase in hypernatremia prevalence, while the opposite was the case for sodium correction for total protein. The correction for total protein led to a decrease in mean sodium levels overall. This was because total protein levels in at least three-quarters of the study population (based on the 75th percentiles among patients as shown in Fig. 3) were below 7.1 g/dl, where the corrected sodium level became lower than the measured level according to the correction equation for total protein. We observed that glucose correction decreased the prevalence of hyponatremia and increased that of hypernatremia, consistent with findings from a previous study [18]. Moreover, we found that total protein correction increased the prevalence of hyponatremia by 22.2% (from 23.1% to 45.3%), which is consistent with a 27% increase among patients admitted to the critical care unit and a 36% increase among patients receiving parenteral nutrition [17]. The differences in prevalence following correction for total protein in those studies were likely due to the lower total protein levels of their included patients (ranging between 5.1-5.3 g/dl) compared to those in our study (median, 6.5 g/dl). Correction for high triglycerides did not affect the prevalence of hyponatremia and hypernatremia, probably because the correction was performed in only 361 and 71 individuals, respectively. Our study is novel in that we corrected for each value among overall hospitalized patients (i.e. not limited to patients admitted to intensive care unit), highlighting that when physicians assess the presence of hyponatremia or hypernatremia, they can easily misclassify patients if they do not consider glucose or total protein levels. This potential for misclassification can lead to underreporting of both conditions in clinical practice.
This study had a few key limitations that are worth noting. First, it was performed on data from inpatient clinical settings. The insights obtained may not be generalizable to outpatient settings. Second, we did not consider etiologies or clinical circumstances (e.g. surgeries, treatments, or the administration of fluids) that may have induced hyponatremia or hypernatremia. These types of intervention may affect disease prevalence and diagnosis frequency. In addition, we do not have access to data on sodium levels measured during blood gas testing, which usually uses direct measurement without diluting samples [47]. In this method, we can measure the osmolality gap [11], and the measured sodium concentrations are less likely to be affected by solutes [12]. Therefore, solute-corrected concentrations may not have been as accurate as concentrations measured using direct ion-specific electrodes. At the same time, it should require attention in interpreting the results of this study, in which data were used without confirming the sodium levels with direct ion-specific electrode measurements. This is because we had no access to the methods of sodium measurement used in this study. Finally, we used data from DPC hospitals that submitted laboratory and claims data to JMDC Inc. These hospitals may not represent all DPC hospitals in Japan.
This retrospective validation study revealed that ICD-10-based diagnoses of both hyponatremia and hypernatremia had low sensitivity and high specificity, suggesting that recorded hyponatremia and hypernatremia should not be used for calculations of prevalence or incidence but may be used for relative risk calculations. In addition, our analyses also suggest that the failure to adjust for blood glucose, triglycerides, or total protein may underestimate or overestimate the prevalence of both sodium disorders.
ACKNOWLEDGEMENTS
A.O., S.Y., K.I.K., R.I., and T.K. are members of the Department of Prevention of Diabetes and Lifestyle-Related Diseases, which is a cooperative program between the University of Tokyo and Asahi Mutual Life Insurance Company. K.I.K. was previously employed by the Asahi Mutual Life Insurance Company. The content is the sole responsibility of the authors and does not necessarily represent the official views of the University of Tokyo. The other authors declare no potential conflicts of interest.
AUTHORS’ CONTRIBUTIONS
Study design, A.O. and H.Y.; data acquisition, A.O.; statistical analyses, A.O., H.Y., S.O., S.Y., and H.Y.; data interpretation, H.W., K.M., S.O., K.I.K., M.N., M.K., R.I., H.Y., T.Y., T.K., S.Y., and M.N.; supervision, H.Y., T.Y., T.K., S.Y., and M.N.; funding acquisition, A.O. Each author contributed important intellectual content during manuscript drafting or revision, agreed to be personally accountable for the individual’s own contributions, and ensured that questions pertaining to the accuracy or integrity of any portion of the work, even one in which the author was not directly involved, were appropriately investigated and resolved, including documentation in the literature, if appropriate.
FUNDING
This work was supported by grants from the Ministry of Education, Culture, Sports, Science and Technology, Japan (20K18957); the Kidney Foundation of Japan (JKFB 20-8); the Japanese Association of Dialysis Physicians (JADP grant 2020‐4); and a junior scientist development grant from the Japan Diabetes Society to A.O. The funding sources were not involved in the design or conduct of the study; the collection, management, analysis, or interpretation of the data; the preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.
DATA AVAILABILITY STATEMENT
The data underlying this article were provided by JMDC Inc. under license. Data will be shared on request to the corresponding author with permission of JMDC Inc.
CONFLICT OF INTEREST STATEMENT
None declared.
REFERENCES
Nguồn: https://buycookiesonline.eu
Danh mục: Info
