Abstract
Background Hospitalizations and 30-day readmissions are common in the hemodialysis population. Actionable clinical markers for near-term hospital encounters are needed to identify individuals who require swift intervention to avoid hospitalization. Aspects of volume management, such as failed target weight (i.e, estimated dry weight) achievement, are plausible modifiable indicators of impending adverse events. The short-term consequences of failed target weight achievement are not well established.
Methods Statistically deidentified data were taken from a cohort of Medicare-enrolled, prevalent hemodialysis patients treated at a large dialysis organization from 2010 to 2012. We used a retrospective cohort design with repeated intervals, each consisting of 180-day baseline, 30-day exposure assessment, and 30-day follow-up period, to estimate the associations between failed target weight achievement and the risk of 30-day emergency department visits and hospitalizations. We estimated adjusted risk differences using inverse probability of exposure weighted Kaplan–Meier methods.
Results A total of 113,561 patients on hemodialysis contributed 788,722 study intervals to analyses. Patients who had a postdialysis weight >1.0 kg above the prescribed target weight in ≥30% (versus <30%) of exposure period treatments had a higher absolute risk (risk difference) of 30-day: emergency department visits (2.13%; 95% confidence interval, 2.00% to 2.32%); and all-cause (1.47%; 95% confidence interval, 1.34% to 1.62%), cardiovascular (0.31%; 95% confidence interval, 0.24% to 0.40%), and volume-related (0.15%; 95% confidence interval, 0.11% to 0.21%) hospitalizations.
Conclusions In the absence of objective measures of volume status, recurrent failure to achieve target weight is an easily identifiable clinical risk marker for impending hospital encounters among patients on hemodialysis.
Keywords: ED visits, hemodialysis, hospitalizations, target weight
Individuals receiving maintenance hemodialysis in the United States have unacceptably high hospitalization and 30-day readmission rates.1 Easily discernible and actionable clinical markers for near-term hospital encounters are needed to identify individuals who require swift intervention to avoid hospitalization. Many patient-related hospitalization risk factors, such as heart failure, catheter vascular access, and hypoalbuminemia, are difficult to modify quickly. In contrast, dialysis treatment–related factors, such as aspects of volume management, can often be addressed more promptly. In fact, volume overload is a major contributor to hospitalizations, readmissions, and their associated costs in the hemodialysis population. One report estimated that hospital encounters for volume overload cost Medicare $266 million from 2006 to 2008.2
Both inaccurate prescribed target weight (i.e., estimated dry weight) and failure to achieve prescribed target weight can lead to volume overload. Imprecision in the assessment of volume status and the absence of universally accepted objective volume measurement tools make it challenging to recognize a misestimated target weight. The recurrent occurrence of postdialysis weight exceeding target weight may be an easily identifiable marker of volume overload in some patients. A previous study found that individuals who experienced postdialysis weights >2.0 kg above the prescribed target weight (versus those with postdialysis weights within 2.0 kg of target weight) had a greater long-term mortality risk.3 However, the short-term consequences of failed target weight achievement are not well established. In addition, some experts have suggested that attention to and adjustment of prescribed target weight after hospital discharge may reduce readmission risk,4,5 but existing data do not support this seemingly sound hypothesis.
We undertook this study to investigate the association between failed target weight achievement and the short-term risk of emergency department (ED) visits and inpatient hospitalizations in a large, nationally representative cohort of prevalent hemodialysis patients from the United States, using a modern epidemiologic study design and analytic methods. In post hoc analyses, we evaluated the association between posthospitalization target weight adjustment and the occurrence of short-term outcomes, including 30-day hospital readmissions.
Methods
This study was approved by the University of North Carolina at Chapel Hill Institutional Review Board (17–0011). Data from a large, for-profit dialysis organization with over 1500 outpatient dialysis clinics located across the United States were linked with the US Renal Data System at the patient level.
Study Design and Population
We used a retrospective cohort design with repeated intervals, each consisting of a 180-day baseline, a 30-day exposure assessment, and a 30-day follow-up period, to investigate the associations between postdialysis weight above the prescribed target weight and short-term clinical outcomes among individuals receiving maintenance hemodialysis at the dialysis organization from 2010 to 2012. Study intervals were indexed to equilibrated kt/v (eKt/V) measurement dates to promote consistency with the planned Centers of Medicare and Medicaid ESRD Quality Incentive Program ultrafiltration rate reporting measure.6 Patients entered the cohort at the time of their first eligible eKt/V measurement during the study period, and subsequent intervals were indexed to later eKt/V measurements. Individual study intervals were constructed, such that (1) the exposure period began on the day after the indexing eKt/V measurement, (2) the follow-up period began immediately after the end of the exposure period, and (3) the baseline period for covariate ascertainment began 180 days before the start of the exposure period (Figure 1A). Within individuals, all exposure/follow-up periods occurring across time were discrete (i.e., did not overlap). However, baseline periods of study intervals occurring later in time could overlap with prior intervals as long as their associated exposure/follow-up periods did not overlap (Supplemental Figure 1).7
Figure 1.
Study designs. (A) In primary and secondary analyses, a retrospective cohort design with repeated intervals, each consisting of a 180-day baseline, a 30-day exposure assessment and a 30-day follow-up period was used. Exposure periods were anchored on eKt/V measurements occurring during the study period. This figure illustrates a single study interval; individuals could contribute multiple study intervals to analyses. Individual study intervals were constructed such that the: 1) exposure period began on the day after the indexing eKt/V measurement; 2) follow-up period began immediately after the end of the exposure period; and 3) baseline period for covariate ascertainment began 180 days prior to the start of the exposure period. (B) The study design for post hoc analyses involved identifying hospitalization events that occurred during 30-day follow-up intervals in the primary study. Exposure periods were anchored on hospital discharge dates. This figure illustrates a single study interval; individuals could contribute multiple study intervals to analyses. For each eligible hospitalization event the: 1) 7-day exposure period began on the day following hospital discharge; and 2) the follow-up period began immediately after the end of the exposure period and continued up to day 30 post-hospitalization. eKt/V, equilibrated kt/v; Rx, prescribed.
The study population was composed of Medicare-enrolled, prevalent hemodialysis patients who had at least one eKt/V measurement between January 1, 2010 and December 31, 2012. We studied a prevalent population to minimize confounding from residual kidney function and other dynamic clinical features in the first year of dialysis. Among eligible individuals, baseline/exposure/follow-up intervals were excluded if patients (1) had a dialysis vintage <1 year at the time of the index eKt/V measurement, (2) were <18 years of age, (3) received home hemodialysis or peritoneal dialysis during the baseline or exposure period, (4) received fewer than six center-based hemodialysis treatments during the exposure period, (5) did not have continuous Medicare Part A/B coverage during the baseline and exposure periods, (6) experienced frequent postdialysis weight below the prescribed target weight during the exposure period (postdialysis weight >1.0 kg below target weight in ≥30% of treatments),3 or (7) had a hospital admission that extended from the exposure period into the follow-up period. Additionally, within individual patients, potential study intervals occurring later in time were excluded if their associated exposure/follow-up periods would overlap with the exposure/follow-up period of a previously included interval.
Study Exposure, Outcomes, and Covariates
The exposure of interest was postdialysis weight above the prescribed target weight. In primary analyses, patients were classified as failing to achieve target weight if their postdialysis weight was >1.0 kg above their prescribed target weight in ≥30% of exposure period hemodialysis treatments. Patients with a postdialysis weight >1.0 kg above their target weight in <30% of exposure period treatments served as the referent group. In secondary analyses, we considered (1) alternative binary exposures using different weight thresholds, including postdialysis weights >1.5, >2.0, and >2.5 kg above the prescribed target weight in ≥30% versus <30% of exposure period treatments; (2) a multilevel categorical exposure parameterized as <5% (referent), 5%–29%, 30%–49%, and ≥50% of exposure period treatments with a postdialysis weight >1.0 kg above the prescribed target weight; and (3) a mean-based exposure comparing individuals with an average postdialysis weight minus prescribed target weight of >1.0 versus ≤1.0 kg during the exposure period.
Our primary study outcomes included short-term (7-, 14-, and 30-day) (1) ED visits8 and (2) inpatient hospitalizations (all-cause, cardiovascular, and volume related).9,10 Secondary outcomes included short-term all-cause and cardiovascular mortality.9 All study outcomes were assessed separately, and their definitions are provided in Supplemental Table 1.
Covariates were identified in each eligible 180-day baseline period and included demographic characteristics, comorbid conditions, laboratory data, dialysis treatment parameters, and metrics of health care utilization (Supplemental Table 2). Missing baseline laboratory variables were imputed using the Markov Chain Monte Carlo method with ten imputations.11 All baseline covariates were included in the imputation.
Statistical Analyses
All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC). Baseline characteristics were described across individuals who did and did not experience frequent postdialysis weight >1.0 kg above prescribed target weight as count (percentage) for categorical variables and mean±SD for continuous variables. Baseline covariate distributions were compared using standardized differences. A standardized difference >10.0% represents an imbalance between exposure groups.12
Across all analyses, we used Kaplan–Meier methods to estimate (1) risk differences (RDs), an absolute effect measure (null value =0.00%); and (2) risk ratios, a relative effect measure (null value =1.00).7,13 To assess the relationship between failure to achieve target weight and the risk of short-term ED visits and hospitalizations, we estimated the 30-day cumulative incidence function of each outcome within each exposure group using the complement of the Kaplan–Meier estimator. We then estimated 7-, 14-, and 30-day RDs and risk ratios from these cumulative incidence functions. To account for the within-person correlation of the repeated measures (i.e., multiple intervals per patient), we obtained 95% confidence intervals (95% CIs) using a cluster-based bootstrap procedure14 on the basis of 250 resamples. Inverse probability of exposure (IPE) weighting was used for confounding control. We used multivariable logistic regression to calculate the predicted probability (i.e., propensity score) of exposure as a function of baseline covariates. Propensity scores were then used to generate IPE weights.15 Statistical adjustment was performed by applying IPE weights to the Kaplan–Meier estimator. Across each study interval, patients surviving the exposure period were followed forward in historical time to the first occurrence of a study outcome or censoring event. Censoring events included (1) kidney transplantation, (2) dialysis modality change, (3) recovery of kidney function, (4) loss of Medicare Part A/B coverage, (4) lost to follow-up, and (5) end of the specified follow-up period (7, 14, or 30 days). In hospitalization analyses, death was treated as an additional censoring event. We also calculated incidence rate differences per 100 person-years, an absolute effect measure (null value =0.00), by subtracting outcome rates in the (−) failure to achieve target weight group from outcome rates in the (+) failure to achieve target weight group.16 Finally, to assess the robustness of our primary study findings, we evaluated the failed target weight achievement-hospital encounter associations among individuals with and without exposure period target weight adjustments.
Post Hoc Analyses
We conducted post hoc analyses to explore the potential benefit of adjusting prescribed target weight immediately after inpatient hospitalizations. Among individuals who experienced an all-cause hospitalization during follow-up in our primary study, we evaluated the association between prescribed target weight adjustment within 7 days of hospital discharge (versus not) and the occurrence of 30-day ED visits, hospital readmissions, and a composite outcome of an ED visit, hospitalization, or death (Figure 1B, Supplemental Figure 2). For each eligible hospitalization, the day after hospital discharge was designated as the start of the 7-day exposure assessment period. Study follow-up began immediately after the end of the exposure period and continued up to day 30 posthospitalization. In addition to the covariates used in primary analyses, we adjusted post hoc analyses for length of hospital stay and primary hospitalization cause (cardiovascular, infectious, or other).
Results
Study Cohort Characteristics
Figure 2 displays a flow diagram of study cohort selection. A total of 113,561 patients on hemodialysis contributed 788,722 intervals (7.0±4.5 intervals per patient) to analyses. Individuals underwent an average of 12.4±1.5 treatments during exposure assessment periods. Overall, study patients had an average age of 61.1±14.8 years, 45.1% were female, 42.4% were black, and the most common ESRD cause was diabetes (44.5%).
Figure 2.
Flow diagram depicting the assembly of the primary cohort. eKt/V, equilibrated kt/v.
Baseline characteristics of the primary cohort stratified by target weight achievement status (postdialysis weight >1.0 kg above the prescribed target weight in ≥30% and <30% of exposure period treatments) are presented in Table 1. Individuals who experienced postdialysis weights above the prescribed target weight more (versus less) frequently during exposure periods were younger and had larger body sizes, a greater cardiovascular disease burden, shorter delivered dialysis treatment times, larger interdialytic weight gains, and more missed outpatient hemodialysis treatments. After IPE weighting, all baseline covariates were well balanced between exposure groups (standardized differences <10.0%).
Table 1.
Baseline characteristics of individuals experiencing postdialysis weight >1.0 kg above prescribed target weight in ≥30% versus <30% of exposure period treatments
| Characteristic | Unweighted | Weighted | ||||
|---|---|---|---|---|---|---|
| Postdialysis Weight >1.0 kg above Target Weight during the Exposure Period | Standardized Difference, %a | Postdialysis Weight >1.0 kg above Target Weight during the Exposure Period | Standardized Difference, %a | |||
| ≥30% of Treatments, n=234,102 | <30% of Treatments, n=554,620 | ≥30% of Treatments, n=232,716 | <30% of Treatments, n=555,520 | |||
| Age, yr | 59.0±14.6 | 62.0±14.8 | 20.7 | 61.2±14.8 | 61.2±14.8 | 0.1 |
| Women | 107,579 (46.0) | 248,399 (44.8) | 2.3 | 105,163 (45.2) | 251,151 (45.2) | 0.0 |
| Race | 6.2 | 0.6 | ||||
| White | 117,371 (50.1) | 292,493 (52.7) | 120,559 (51.8) | 288,521 (51.9) | ||
| Black | 104,189 (44.5) | 229,884 (41.4) | 99,217 (42.6) | 235,559 (42.4) | ||
| Other | 12,542 (5.4) | 32,243 (5.8) | 12,939 (5.6) | 31,439 (5.7) | ||
| Hispanic | 40,066 (17.1) | 95,570 (17.2) | 0.3 | 39,839 (17.1) | 95,377 (17.2) | 0.1 |
| Cause of ESRD | 13.7 | 0.3 | ||||
| Diabetes | 115,433 (49.3) | 235,842 (42.5) | 104,082 (44.7) | 247,710 (44.6) | ||
| Hypertension | 63,533 (27.1) | 170,722 (30.8) | 68,885 (29.6) | 164,849 (29.7) | ||
| Glomerular disease | 24,977 (10.7) | 66,229 (11.9) | 26,766 (11.5) | 64,116 (11.5) | ||
| Other | 30,159 (12.9) | 81,827 (14.8) | 32,982 (14.2) | 78,845 (14.2) | ||
| Dialysis vintage, yr | 4.8 | 0.6 | ||||
| 1.0–1.9 | 46,772 (20.0) | 100,705 (18.2) | 43,163 (18.5) | 103,799 (18.7) | ||
| 2.0–3.9 | 67,688 (28.9) | 160,790 (29.0) | 66,970 (28.8) | 160,774 (28.9) | ||
| ≥4.0 | 119,642 (51.1) | 293,125 (52.9) | 122,583 (52.7) | 290,947 (52.4) | ||
| History of a prior kidney transplant | 16,581 (7.1) | 41,924 (7.6) | 1.8 | 17,278 (7.4) | 41,166 (7.4) | 0.1 |
| Body mass index, kg/m2 | 25.3 | 0.4 | ||||
| <18.5 | 6815 (2.9) | 20,905 (3.8) | 8337 (3.6) | 19,646 (3.5) | ||
| 18.5–24.9 | 67,713 (28.9) | 205,804 (37.1) | 81,127 (34.9) | 192,824 (34.7) | ||
| 25.0–29.9 | 63,469 (27.1) | 164,421 (29.6) | 66,961 (28.8) | 160,337 (28.9) | ||
| ≥30.0 | 96,105 (41.1) | 163,490 (29.5) | 76,292 (32.8) | 182,713 (32.9) | ||
| Ischemic heart disease | 75,683 (32.3) | 150,146 (27.1) | 11.5 | 67,390 (29.0) | 159,724 (28.8) | 0.5 |
| Heart failure | 77,476 (33.1) | 129,440 (23.3) | 21.8 | 62,159 (26.7) | 146,591 (26.4) | 0.7 |
| Hypertension | 170,828 (73.0) | 349,941 (63.1) | 21.3 | 154,740 (66.5) | 367,449 (66.1) | 0.7 |
| Diabetes | 155,491 (66.4) | 322,189 (58.1) | 17.2 | 141,610 (60.9) | 336,892 (60.6) | 0.4 |
| Malignancy | 18,084 (7.7) | 47,263 (8.5) | 2.9 | 19,515 (8.4) | 46,095 (8.3) | 0.3 |
| History of noncomplianceb | 13,099 (5.6) | 13,855 (2.5) | 15.8 | 8107 (3.5) | 19,166 (3.5) | 0.2 |
| Albumin,c g/dl | 16.5 | 0.8 | ||||
| ≤3.0 | 8594 (3.7) | 11,675 (2.1) | 6251 (2.7) | 14,637 (2.6) | ||
| 3.1–3.5 | 32,504 (13.9) | 59,112 (10.7) | 27,731 (11.9) | 65,147 (11.7) | ||
| 3.6–4.0 | 110,949 (47.4) | 255,785 (46.1) | 108,276 (46.5) | 258,059 (46.5) | ||
| ≥4.1 | 82,055 (35.1) | 228,048 (41.1) | 90,458 (38.9) | 217,676 (39.2) | ||
| Calcium,c mg/dl | 10.4 | 0.4 | ||||
| ≤8.4 | 43,847 (18.7) | 82,481 (14.9) | 37,160 (16.0) | 88,968 (16.0) | ||
| 8.5–10.2 | 181,612 (77.6) | 449,080 (81.0) | 186,020 (79.9) | 444,156 (80.0) | ||
| ≥10.3 | 8643 (3.7) | 23,059 (4.2) | 9535 (4.1) | 22,396 (4.0) | ||
| Phosphorus,c mg/dl | 22.2 | 0.6 | ||||
| ≤3.4 | 21,073 (9.0) | 62,677 (11.3) | 24,637 (10.6) | 59,030 (10.6) | ||
| 3.5–5.5 | 114,487 (48.9) | 317,257 (57.2) | 126,630 (54.4) | 303,638 (54.7) | ||
| ≥5.6 | 98,542 (42.1) | 174,686 (31.5) | 81,449 (35.0) | 192,851 (34.7) | ||
| Potassium,c mEq/L | 14.6 | 0.2 | ||||
| ≤3.9 | 16,506 (7.1) | 51,316 (9.3) | 20,011 (8.6) | 47,772 (8.6) | ||
| 4.0–6.0 | 203,915 (87.1) | 485,456 (87.5) | 203,241 (87.3) | 485,333 (87.4) | ||
| ≥6.1 | 13,681 (5.8) | 17,848 (3.2) | 9464 (4.1) | 22,414 (4.0) | ||
| Hemoglobin,c g/dl | 8.2 | 0.4 | ||||
| ≤9.4 | 18,333 (7.8) | 32,057 (5.8) | 15,142 (6.5) | 35,757 (6.4) | ||
| 9.5–11.9 | 158,480 (67.7) | 381,544 (68.8) | 158,743 (68.2) | 380,010 (68.4) | ||
| ≥12 | 57,289 (24.5) | 141,019 (25.4) | 58,831 (25.3) | 139,753 (25.2) | ||
| Transferrin saturation,c % | 4.2 | 0.3 | ||||
| ≤19 | 39,215 (16.8) | 85,039 (15.3) | 36,908 (15.9) | 87,728 (15.8) | ||
| 20–29 | 88,517 (37.8) | 209,354 (37.7) | 87,964 (37.8) | 209,720 (37.8) | ||
| ≥30 | 106,370 (45.4) | 260,227 (46.9) | 107,843 (46.3) | 258,071 (46.5) | ||
| Creatinine,c mg/dl | 4.8 | 0.3 | ||||
| ≤6.7 | 56,706 (24.2) | 140,693 (25.4) | 58,079 (25.0) | 139,287 (25.1) | ||
| 6.8–8.5 | 57,280 (24.5) | 140,693 (25.4) | 58,454 (25.1) | 139,510 (25.1) | ||
| 8.6–10.6 | 58,604 (25.0) | 138,145 (24.9) | 58,111 (25.0) | 138,508 (24.9) | ||
| ≥10.7 | 61,512 (26.3) | 135,089 (24.4) | 58,071 (25.0) | 138,215 (24.9) | ||
| eKt/V<1.2 | 3743 (1.6) | 4113 (0.7) | 8.0 | 2361 (1.0) | 5638 (1.0) | 0.0 |
| Vascular access | 11.9 | 0.6 | ||||
| Fistula | 145,596 (62.2) | 363,740 (65.6) | 149,365 (64.2) | 357,969 (64.4) | ||
| Graft | 56,877 (24.3) | 136,849 (24.7) | 57,400 (24.7) | 136,497 (24.6) | ||
| Catheter | 31,629 (13.5) | 54,031 (9.7) | 25,950 (11.2) | 61,053 (11.0) | ||
| Treatment time ≥240 min | 50,415 (21.5) | 97,027 (17.5) | 10.2 | 43,344 (18.6) | 103,870 (18.7) | 0.6 |
| Interdialytic weight gain, kg | 57.3 | 0.0 | ||||
| ≤1.9 | 32,338 (13.8) | 164,915 (29.7) | 57,493 (24.7) | 138,637 (25.0) | ||
| 2.0–2.5 | 45,576 (19.5) | 151,701 (27.4) | 58,241 (25.0) | 138,805 (25.0) | ||
| 2.6–3.3 | 62,187 (26.6) | 134,825 (24.3) | 58,436 (25.1) | 138,780 (25.0) | ||
| ≥3.4 | 94,001 (40.2) | 103,179 (18.6) | 58,545 (25.2) | 139,298 (25.1) | ||
| Postdialysis weight, kg | 27.4 | 0.0 | ||||
| ≤64.3 | 47,813 (20.4) | 149,380 (26.9) | 58,587 (25.2) | 139,242 (25.1) | ||
| 64.4–76.4 | 51,282 (21.9) | 145,886 (26.3) | 58,021 (24.9) | 138,886 (25.0) | ||
| 76.5–92.0 | 57,712 (24.7) | 139,472 (25.1) | 58,254 (25.0) | 138,858 (25.0) | ||
| ≥92.1 | 77,295 (33.0) | 119,882 (21.6) | 57,854 (24.9) | 138,534 (24.9) | ||
| Predialysis systolic BP, mm Hg | 7.7 | 0.0 | ||||
| ≤129 | 42,599 (18.2) | 96,055 (17.3) | 41,789 (18.0) | 98,162 (17.7) | ||
| 130–149 | 76,266 (32.6) | 192,410 (34.7) | 78,824 (33.9) | 188,988 (34.0) | ||
| 150–169 | 74,252 (31.7) | 182,660 (32.9) | 75,250 (32.3) | 180,618 (32.5) | ||
| ≥170 | 40,985 (17.5) | 83,495 (15.1) | 36,853 (15.8) | 87,752 (15.8) | ||
| No. of hospital admissionsd | 30.8 | 0.0 | ||||
| 0 | 121,583 (51.9) | 362,690 (65.4) | 141,050 (60.6) | 339,835 (61.2) | ||
| 1–2 | 79,213 (33.8) | 153,143 (27.6) | 69,734 (30.0) | 164,024 (29.5) | ||
| ≥3 | 33,306 (14.2) | 38,787 (7.0) | 21,932 (9.4) | 51,661 (9.3) | ||
| No. of missed dialysis treatments (unexcused)e | 10.4 | 0.4 | ||||
| 0 | 204,900 (87.5) | 503,016 (90.7) | 208,431 (89.6) | 498,294 (89.7) | ||
| 1–2 | 18,551 (7.9) | 31,374 (5.7) | 14,984 (6.4) | 35,347 (6.4) | ||
| ≥3 | 10,651 (4.5) | 20,230 (3.6) | 9301 (4.0) | 21,879 (3.9) | ||
A total of 113,561 unique patients on hemodialysis contributed 788,722 intervals to the analysis. In each study interval, covariates were measured during the 180-day baseline period before the 30-day exposure period. Values are given as count (%) for categorical variables and mean±SD for continuous variables. eKt/V, equilibrated kt/v; No., number.
A standardized difference >10.0% represents meaningful imbalance between groups.
The claims-based definition of noncompliance included International Classification of Diseases, Ninth Revision discharge diagnosis code V15.81 (personal history of noncompliance with medical treatment, presenting hazards to health).
Values were imputed using the Markov Chain Monte Carlo method using ten imputations when missing (albumin, n=216; calcium, n=216; creatinine, n=26,787; hemoglobin, n=13; phosphorus, n=140; potassium, n=214; and transferrin saturation, n=1059).11
Number of inpatient hospitalizations during the entire 180-day baseline period.
Number of missed hemodialysis treatments (unexcused) in the last 30 days of the 180-day baseline period.
Primary Analyses
In primary analyses, we evaluated the associations between failed target weight achievement and 30-day hospital encounters using a binary exposure variable. During the exposure period, the median [quartile 1, quartile 3] postdialysis weight minus target weight was 1.3 [0.5, 2.4] kg among patients who had a postdialysis weight >1.0 kg above target weight in ≥30% compared with 0.0 [−0.4, 0.4] kg among patients who had a postdialysis weight >1.0 kg above target weight in <30% exposure period treatments. Experiencing a postdialysis weight >1.0 kg above target weight in ≥30% (versus <30%) of exposure period treatments was associated with higher 7-, 14-, and 30-day risks of ED visits and hospitalizations. The 30-day adjusted RDs (95% CIs) were 2.13% (95% CI, 2.00% to 2.32%) for ED visits and 1.47% (95% CI, 1.34% to 1.62%) for all-cause hospitalizations. Thus, failing to achieve target weight (versus not) was associated with 21 excess ED visits and 15 excess all-cause hospitalizations per 1000 persons during the 30-day follow-up period. Analyses considering cause-specific hospitalizations (cardiovascular and volume related) yielded similar results (Supplemental Table 3, Table 2).
Table 2.
Associations between postdialysis weight >1.0 kg above target weight in ≥30% versus <30% of exposure period treatments and 7-, 14-, and 30-day hospital encounters
| Outcome | Postdialysis Weight >1.0 kg above Target Weight during the Exposure Period in ≥30% versus <30% of Treatments (ref.) | ||
|---|---|---|---|
| Adjusted Risk Difference, % (95% CI) | Adjusted Incidence Rate Difference per 100 person-yr (95% CI) | Adjusted Risk Ratio (95% CI) | |
| All-cause ED visits | |||
| 7-d Follow-up | 0.71 (0.61 to 0.82) | 38.82 (33.68 to 44.86) | 1.14 (1.12 to 1.17) |
| 14-d Follow-up | 1.24 (1.12 to 1.37) | 35.62 (32.25 to 39.47) | 1.14 (1.12 to 1.15) |
| 30-d Follow-up | 2.13 (2.00 to 2.32) | 31.74 (29.76 to 34.53) | 1.13 (1.12 to 1.14) |
| All-cause hospitalizations | |||
| 7-d Follow-up | 0.43 (0.35 to 0.52) | 23.19 (18.85 to 27.60) | 1.14 (1.11 to 1.17) |
| 14-d Follow-up | 0.85 (0.73 to 0.95) | 23.43 (20.27 to 26.38) | 1.14 (1.13 to 1.16) |
| 30-d Follow-up | 1.47 (1.34 to 1.62) | 20.46 (18.64 to 22.45) | 1.13 (1.12 to 1.14) |
| Cardiovascular hospitalizations | |||
| 7-d Follow-up | 0.11 (0.07 to 0.15) | 5.70 (3.60 to 7.91) | 1.12 (1.07 to 1.17) |
| 14-d Follow-up | 0.17 (0.12 to 0.24) | 4.61 (3.24 to 6.45) | 1.10 (1.07 to 1.14) |
| 30-d Follow-up | 0.31 (0.24 to 0.40) | 3.95 (3.07 to 5.05) | 1.09 (1.07 to 1.11) |
| Volume-related hospitalizations | |||
| 7-d Follow-up | 0.06 (0.04 to 0.09) | 3.11 (1.91 to 4.48) | 1.17 (1.10 to 1.25) |
| 14-d Follow-up | 0.07 (0.04 to 0.11) | 1.90 (1.00 to 2.92) | 1.11 (1.05 to 1.17) |
| 30-d Follow-up | 0.15 (0.11 to 0.21) | 1.90 (1.36 to 2.61) | 1.11 (1.08 to 1.15) |
Adjusted risk differences are percentages (%). The null value for a risk difference is 0.00%, the null value for an incidence rate difference is 0.00, and the null value for a risk ratio is 1.00. 95% CI, 95% confidence interval; ED, emergency department; ref., referent.
Prescribed target weight was adjusted in 277,294 (35.2%) exposure periods. A total of 136,776 exposure periods had at least one downward target weight adjustment, and a total of 164,488 exposure periods had at least one upward adjustment. Across all study intervals, the median [quartile 1, quartile 3] number of prescribed target weight adjustments during exposure periods was 0 [0, 1]. Sensitivity analyses considering individuals with and without changes in prescribed target weight (separately) during the exposure period generated results similar to those of our primary analyses (Supplemental Table 4).
Secondary Analyses
In secondary analyses, we evaluated the associations between failed target weight achievement and 30-day hospital encounters using alternative exposure specifications. First, we varied the weight threshold used to define postdialysis weight above the prescribed target weight (>1.5, >2.0, and >2.5 kg). As the weight threshold for exposure classification increased, the magnitude of the observed associations between postdialysis weight above the prescribed target weight and short-term hospital encounters also increased (Figure 3, Supplemental Table 5). Second, we considered failed target weight achievement as a multilevel categorical variable (<5% [referent], 5%–29%, 30%–49%, and ≥50% of exposure period dialysis treatments with a postdialysis weight >1.0 kg above the prescribed target weight). Failing to achieve prescribed target weight on a more frequent basis was associated with an incrementally greater risk of 30-day ED visits and hospitalizations (Table 3, Supplemental Table 6). Third, we assessed the association between having an average postdialysis weight minus target weight of >1.0 kg (versus ≤1.0 kg) during the exposure period and 30-day hospital encounters. Results were analogous to primary study findings (Supplemental Table 7).
Figure 3.
Associations between postdialysis weight above target weight in ≥30% versus <30% of exposure period treatments and 30-day all-cause hospital encounters across varying kilogram thresholds of failed target weight achievement. (A) Depicts the association between post-dialysis weight above the prescribed target weight in ≥30% versus <30% of exposure period treatments and 30-day all-cause ED visits across varying kilogram thresholds of failed target weight achievement. (B) Depicts the association between post-dialysis weight above the prescribed target weight in ≥30% versus <30% of exposure period treatments and 30-day all-cause hospitalizations across varying kilogram thresholds of failed target weight achievement. The null value for a risk difference is 0.00%. As the weight threshold for failure to attain target weight increased, the magnitude of the observed associations between postdialysis weight above the prescribed target weight and short-term hospital encounters also increased. ED, emergency department; ref., referent; wt, weight.
Table 3.
Associations between postdialysis weight >1.0 kg above target weight during the exposure period and 30-day hospital encounters using varied frequency thresholds for failed target weight achievement
| Outcome | Postdialysis Weight >1.0 kg above Target Weight during the Exposure Period | |||
|---|---|---|---|---|
| <5% of Treatments | 5%–29% of Treatments | 30%–49% of Treatments | ≥50% of Treatments | |
| All-cause ED visits | ||||
| Adjusted risk difference, % (95% CI) | 0.00 (ref.) | 1.19 (1.00 to 1.37) | 1.98 (1.73 to 2.24) | 3.07 (2.82 to 3.34) |
| Adjusted incidence rate difference (95% CI) | 0.00 (ref.) | 17.52 (14.86 to 20.25) | 29.04 (25.71 to 33.12) | 45.96 (42.66 to 49.99) |
| Adjusted risk ratio (95% CI) | 1.00 (ref.) | 1.07 (1.06 to 1.09) | 1.12 (1.11 to 1.14) | 1.19 (1.17 to 1.21) |
| All-cause hospitalizations | ||||
| Adjusted risk difference, % (95% CI) | 0.00 (ref.) | 0.69 (0.52 to 0.83) | 1.20 (0.98 to 1.41) | 2.08 (1.90 to 2.29) |
| Adjusted incidence rate difference (95% CI) | 0.00 (ref.) | 9.41 (7.13 to 11.60) | 16.45 (13.80 to 19.20) | 28.89 (25.83 to 31.91) |
| Adjusted risk ratio (95% CI) | 1.00 (ref.) | 1.06 (1.05 to 1.08) | 1.11 (1.09 to 1.13) | 1.19 (1.17 to 1.21) |
| Cardiovascular hospitalizations | ||||
| Adjusted risk difference, % (95% CI) | 0.00 (ref.) | 0.16 (0.04 to 0.23) | 0.21 (0.09 to 0.35) | 0. 45 (0.33 to 0.58) |
| Adjusted incidence rate difference (95% CI) | 0.00 (ref.) | 1.97 (0.81 to 3.16) | 2.64 (0.93 to 4.34) | 5.70 (4.38 to 7.18) |
| Adjusted risk ratio (95% CI) | 1.00 (ref.) | 1.04 (1.01 to 1.07) | 1.06 (1.03 to 1.10) | 1.13 (1.09 to 1.16) |
| Volume-related hospitalizations | ||||
| Adjusted risk difference, % (95% CI) | 0.00 (ref.) | 0.00 (−0.07 to 0.06) | 0.03 (−0.05 to 0.11) | 0.21 (0.13 to 0.28) |
| Adjusted incidence rate difference (95% CI) | 0.00 (ref.) | 0.04 (−0.62 to 0.78) | 0.35 (−0.57 to 1.39) | 2.61 (1.68 to 3.64) |
| Adjusted risk ratio (95% CI) | 1.00 (ref.) | 1.00 (0.95 to 1.04) | 1.02 (0.96 to 1.08) | 1.15 (1.09 to 1.21) |
Adjusted risk differences are percentages (%). Adjusted incidence rate differences are per 100 person-yr. The null value for a risk difference is 0.00%, the null value for an incidence rate difference is 0.00, and the null value for a risk ratio is 1.00. ED, emergency department; 95% CI, 95% confidence interval; ref., referent.
Finally, we assessed the association between failed target weight achievement and mortality. Experiencing a postdialysis weight >1.0 kg above target weight in ≥30% (versus <30%) of exposure period treatments was associated with higher short-term all-cause and cardiovascular mortality (Supplemental Table 3).
Post Hoc Analyses
We conducted post hoc analyses to investigate the potential role of prescribed target weight adjustment in modifying posthospitalization outcomes. We examined the association between having (versus not having) a prescribed target weight adjustment within 7 days of hospital discharge and 30-day posthospitalization outcomes among individuals who experienced inpatient hospitalizations in our primary study cohort (44,460 patients on hemodialysis contributed 71,120 posthospitalization intervals). Individuals whose target weight was (versus was not) adjusted within 7 days of hospital discharge had lower risks of 30-day ED visits, all-cause hospitalizations, and the composite outcome of an ED visit, hospitalization, or death (Figure 4, Supplemental Table 8). On the basis of the estimated RDs, we computed the number needed to treat (95% CI) for each outcome. Target weight adjustment within 7 days of hospital discharge in 114 (95% CI, 60 to 512) individuals may prevent one 30-day ED visit postdischarge, 61 (95% CI, 44 to 110) individuals may prevent one 30-day readmission, and 49 (95% CI, 35 to 74) individuals may prevent one 30-day composite outcome event (an ED visit, hospitalization, or death).
Figure 4.
Associations between having versus not having a prescribed target weight adjustment within 7 days of hospital discharge and 30-day posthospitalization outcomes. The null value for a risk difference is 0.00%. Individuals whose target weight was (versus was not) adjusted within 7 days of hospital discharge had lower risks of 30-day ED visits, all-cause hospitalizations, and the composite outcome. 95% CI, 95% confidence interval; ED, emergency department; ref., referent.
Discussion
To our knowledge, this is the first study evaluating the association between failed target weight achievement and short-term clinical outcomes among individuals receiving maintenance hemodialysis. We found that experiencing a postdialysis weight >1.0 kg above the prescribed target weight in ≥30% (versus <30%) of exposure period treatments was associated with higher risks of 7-, 14-, and 30-day ED visits and hospitalizations (all-cause, cardiovascular, and volume related). The observed associations were (1) more potent when we defined failure to achieve target weight using larger postdialysis weight above prescribed target weight thresholds, (2) incremental when different frequencies of failed target weight achievement were considered, and (3) consistent when we considered failed target weight achievement as a mean-based exposure. Finally, our post hoc analyses showed that adjustment of the target weight prescription after hospital discharge was associated with improved 30-day postdischarge outcomes.
Existing observational evidence suggests that failed target weight achievement is associated with increased long-term mortality. In a prospective study of 182 prevalent Italian hemodialysis patients, Movilli et al.17 reported that an average postdialysis weight of ≥0.3 kg (versus <0.3 kg) above prescribed target was associated with greater 3-year all-cause and cardiovascular mortality. In a retrospective cohort study of 10,785 patients on hemodialysis from the United States, Flythe et al.3 found that individuals who experienced postdialysis weights >2.0 kg above prescribed target weight in ≥30% (versus <30%) of dialysis treatments had higher rates of all-cause and cardiovascular mortality over a median follow-up time of 2.1 years. Both of these studies considered long-term mortality outcomes, and neither examined the relationship of failed target weight achievement and subsequent hospital encounters.
To expand on this evidence base, we used a modern epidemiologic study design where individual patients could contribute multiple baseline/exposure/follow-up intervals to analyses and evaluated the association between postdialysis weight above target weight and short-term clinical outcomes. Recently, a National Institutes of Health working group called out the historical underemphasis of short-term clinical risk prediction,18 and others have noted that dynamic clinical factors (e.g., biochemical indices, interdialytic weight gain, and ultrafiltration volume) tend to be stronger predictors of near- versus long-term clinical events.19 In addition, short-term outcome risk prediction is becoming increasingly important in today’s health care environment, where cost containment and population health management are increasingly valued. Health care models, such as ESRD Seamless Care Organizations and other integrated health care groups, strive to reduce costs by identifying and intervening in individuals who have the highest risk for cost-intensive health care services. Reductions in hospitalizations and 30-day readmissions have been a major focus in this effort.20,21 Our findings suggest that dialysis clinic personnel could use frequent failed target weight achievement as a clinical marker to identify individuals at risk for imminent hospital encounters. Our post hoc analyses offer potential support for this premise. Individuals who had their target weights adjusted after hospitalizations were at lower risk for a 30-day hospital readmission compared with individuals who did not. Such easily recognized risk markers for volume-related hospital encounters are especially important given the limited availability of objective volume status measurement tools in United States outpatient dialysis clinics and the substantial costs associated with volume-related hospitalizations.2
Although our findings suggest that frequent failed target weight achievement may be a clinical marker for an impending hospital encounter, we cannot assess whether the observed risk stems from extracellular volume overload, inadequate volume management, or both. Incorporation of objective volume measurement tools into routine clinical practice to help make this distinction is a critical unmet need in hemodialysis patient care. However, in the absence of objective measures, surrogate risk markers may have clinical utility. Occasional postdialysis-target weight mismatch is expected (e.g., in patients with fluctuating health status, and during intentional target weight challenge). Conversely, frequent postdialysis weight-target weight mismatch likely suggests that a clinical or care process problem exists. Typically, risk stratification approaches rely on computationally complex algorithms performed at the centralized, corporate dialysis provider level.20 In contrast, failed target weight achievement is a simple clinical marker that can be easily identified by a range of clinical care team members, including patent care technicians, nurses, dietitians, medical providers, and patients, rendering it a potentially attractive risk stratification tool that can be used to quickly focus attention to patients at imminent risk for hospital encounters.
Our study has many strengths. First, we selected a study design with repeated intervals, each consisting of separate baseline, exposure, and follow-up periods. Repeatedly ascertaining each patient’s baseline characteristics and exposure status across time ensured that the most current clinical information was considered when assessing failed target weight achievement—short-term outcome associations. Utilization of this design also minimizes biases common to longitudinal observational studies, such as time-varying confounding, immortal person time bias, and selection bias. Second, we used a detailed data source that enabled us to account for numerous demographic, clinical, and dialysis treatment–related variables in our analyses. Covariate imbalances between exposure groups were diminished after IPE weighting. Third, we defined our study exposure, postdialysis weight above the prescribed target weight, using several approaches. Regardless of exposure specification, failure to achieve target weight was consistently associated with a higher risk of short-term adverse outcomes.
Our results must be considered in the context of study limitations. First, our study was observational, and it is possible that residual confounding may exist. To minimize confounding from difficult to measure factors, such as ambient health status, we controlled for variables, including albumin, phosphorus, creatinine, and a history of noncompliance, in our analyses. Second, target weights were prescribed by treating nephrologists in the setting of routine clinical practice. The approach to target weight estimation and adjustment may have varied across clinical providers. Related, target weight changes are made in response to a variety of clinical scenarios and likely have varying consequences. Although we did not assess the association between target weight adjustment and outcomes in our primary analyses, the findings from our post hoc analyses suggest that target weight adjustment soon after hospital discharge associates with improved posthospitalization outcomes. Third, comorbid condition designations were based upon the International Classification of Diseases, Ninth Revision diagnosis codes. Comorbidities not requiring a health care encounter during the 180-day baseline period may have been missed. Fourth, our study included adult, center-based hemodialysis patients with a dialytic vintage of 1 year or longer. Our results may not extrapolate to excluded populations, such as incident hemodialysis, home hemodialysis or peritoneal dialysis patients.
In summary, we found that frequent postdialysis weight ≥1.0 kg above the prescribed target weight was associated with short-term ED visits and hospitalizations. We also observed that target weight adjustment in the period immediately after hospital discharge was associated with a reduced risk of 30-day readmissions. Recurrent failed target weight achievement may represent an easily identifiable risk maker for impending hospital encounters among individuals receiving maintenance hemodialysis. Prospective studies testing clinical monitoring programs that use failed target weight achievement as a clinic-based risk stratification tool for the identification of individuals at high risk for impending hospital encounters are warranted.
Disclosures
M.M.A. and J.E.F. have received investigator-initiated research funding from the Renal Research Institute, a subsidiary of Fresenius Medical Care, North America. In the last 2 years, J.E.F. has received consulting fees from Fresenius Medical Care, North America and speaking honoraria from American Renal Associates, the American Society of Nephrology, Baxter, the National Kidney Foundation, and multiple universities. L.W. has no relevant disclosures.
Supplementary Material
Acknowledgments
The authors acknowledge M. Alan Brookhart of the University of North Carolina at Chapel Hill for his guidance on the analytic approach and data presentation.
Each author contributed important intellectual content during manuscript drafting or revision and accepts accountability for the overall work by ensuring that questions pertaining to the accuracy or integrity of any portion of the work are appropriately investigated and resolved.
J.E.F. is supported by grant K23 DK109401 awarded by the National Institute of Diabetes and Digestive and Kidney Diseases of the National Institutes of Health.
Some of the data reported here have been supplied by DaVita Clinical Research, Inc. and were statistically deidentified. DaVita Clinical Research, Inc. had no role in the design or implementation of this study or in the decision to publish. Additionally, some of the data reported here have been provided by the US Renal Data System. The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as official policy or interpretation of the US Government.
Footnotes
Published online ahead of print. Publication date available at www.jasn.org.
This article contains supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2018010004/-/DCSupplemental.
References
- 1.Saran R, Robinson B, Abbott KC, Agodoa LY, Albertus P, Ayanian J, et al.: US Renal Data System 2016 annual data report: Epidemiology of kidney disease in the United States. Am J Kidney Dis 69[Suppl 1]: A7–A8, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Arneson TJ, Liu J, Qiu Y, Gilbertson DT, Foley RN, Collins AJ: Hospital treatment for fluid overload in the Medicare hemodialysis population. Clin J Am Soc Nephrol 5: 1054–1063, 2010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Flythe JE, Kshirsagar AV, Falk RJ, Brunelli SM: Associations of posthemodialysis weights above and below target weight with all-cause and cardiovascular mortality. Clin J Am Soc Nephrol 10: 808–816, 2015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Chan KE, Lazarus JM, Wingard RL, Hakim RM: Association between repeat hospitalization and early intervention in dialysis patients following hospital discharge. Kidney Int 76: 331–341, 2009 [DOI] [PubMed] [Google Scholar]
- 5.Mathew AT, Strippoli GF, Ruospo M, Fishbane S: Reducing hospital readmissions in patients with end-stage kidney disease. Kidney Int 88: 1250–1260, 2015 [DOI] [PubMed] [Google Scholar]
- 6. Centers for Medicare & Medicaid Services (CMS): End-Stage Renal Disease Quality Incentive Program (ESRD QIP) Payment Year (PY) 2020 Final Measure Technical Specifications. 2017. Available at: https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/ESRDQIP/Downloads/PY-2020-Technical-Measure-Specification_Nov-17-.pdf. Accessed March 28, 2018.
- 7.Brookhart MA, Freburger JK, Ellis AR, Winkelmayer WC, Wang L, Kshirsagar AV: Comparative short-term safety of sodium ferric gluconate versus iron sucrose in hemodialysis patients. Am J Kidney Dis 67: 119–127, 2016 [DOI] [PubMed] [Google Scholar]
- 8.Venkatesh AK, Mei H, Kocher KE, Granovsky M, Obermeyer Z, Spatz ES, et al.: Identification of emergency department visits in medicare administrative claims: Approaches and implications. Acad Emerg Med 24: 422–431, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. United States Renal Data System: ESRD Analytic Methods. 2016. Available at: https://www.usrds.org/2016/view/v2_00_appx.aspx. Accessed November 28, 2017.
- 10.Assimon MM, Nguyen T, Katsanos SL, Brunelli SM, Flythe JE: Identification of volume overload hospitalizations among hemodialysis patients using administrative claims: A validation study. BMC Nephrol 17: 173, 2016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Schafer JL: Multiple imputation: A primer. Stat Methods Med Res 8: 3–15, 1999 [DOI] [PubMed] [Google Scholar]
- 12.Austin PC: Using the standardized difference to compare the prevalence of a binary variable between two groups in observational research. Commun Stat Simul Comput 38: 1228–1234, 2009 [Google Scholar]
- 13.Cole SR, Lau B, Eron JJ, Brookhart MA, Kitahata MM, Martin JN, et al.; CNICS Research Network : Estimation of the standardized risk difference and ratio in a competing risks framework: Application to injection drug use and progression to AIDS after initiation of antiretroviral therapy. Am J Epidemiol 181: 238–245, 2015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Cameron AC, Gelbach JB, Miller DL: Bootstrap-based improvements for inference with clustered errors. Rev Econ Stat 90: 414–427, 2008 [Google Scholar]
- 15.Cole SR, Hernán MA: Constructing inverse probability weights for marginal structural models. Am J Epidemiol 168: 656–664, 2008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Rothman KJ, Greenland S, Lash TL: Modern Epidemiology, Philadelphia, Wolters Kluwer Health/Lippincott Williams & Wilkins, 2008 [Google Scholar]
- 17.Movilli E, Camerini C, Gaggia P, Zubani R, Feller P, Poiatti P, et al.: Magnitude of end-dialysis overweight is associated with all-cause and cardiovascular mortality: A 3-year prospective study. Am J Nephrol 37: 370–377, 2013 [DOI] [PubMed] [Google Scholar]
- 18.Eagle KA, Ginsburg GS, Musunuru K, Aird WC, Balaban RS, Bennett SK, et al.: Identifying patients at high risk of a cardiovascular event in the near future: Current status and future directions: Report of a national heart, lung, and blood institute working group. Circulation 121: 1447–1454, 2010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Goldstein BA, Pencina MJ, Montez-Rath ME, Winkelmayer WC: Predicting mortality over different time horizons: Which data elements are needed? J Am Med Inform Assoc 24: 176–181, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Maddux FW, Ketchersid TL: The journey to full health care responsibility for one ESCO provider. Clin J Am Soc Nephrol 12: 2050–2052, 2017 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Wingard RL, McDougall K, Axley B, Howard A, O''Keefe C, Armistead N, et al.: Right TraC™ post-hospitalization care transitions program to reduce readmissions for hemodialysis patients. Am J Nephrol 45: 532–539, 2017 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.