Transfusion Associated Hyperkalemia in Pediatric Population: Analyses for Risk Factors and Recommendations

Abstract

Background:

Transfusion-associated hyperkalemia (TAH) is a potentially life-threatening complication of red blood cell (RBC) transfusion. Previously, we reported features of RBC transfusions from 35 pediatric patients (TAH group) who had hyperkalemia with RBC transfusion in one-year period at 4 facilities. In this study, we used multivariate analyses and artificial intelligence to compare the TAH group to newly collected control group (non-TAH group) to identify factors associated with TAH occurrence.

Study design:

A review of RBC transfusion with TAH were compared to non-TAH group who did not develop TAH with RBC transfusion at each facility during the same one-year period. The non-TAH group included 12 patients each in 5 age groups. Wilcoxon rank-sum tests recursive feature elimination, least absolute shrinkage, and selection operator (LASSO), and other artificial intelligence techniques were employed to identify the most salient features associated with predicting specific clinical outcomes for TAH occurrence.

Results/Findings:

Pre-transfusion creatinine, comorbidities of kidney and/or liver dysfunctions, and total transfused volume within 12 hours (tV-12) per Kg and per estimated total blood volume (eTBV) showed statistically significant differences between TAH and non-TAH groups. Multivariate analysis revealed the biggest factor on TAH occurrence was tV-12/Kg followed by age of RBC units. The thresholds of risks were tV-12/Kg of 30 ml/Kg, tV-12/eTBV of 30%, and RBC unit age of 7.95 days.

Conclusions:

The study findings suggest that the biggest factor on TAH occurrence is tV-12/Kg. More importantly, 30% of eTBV transfusion could cause TAH in the patients with multiple comorbidities.

Background

Transfusion-associated hyperkalemia (TAH) is a rare but potentially life-threatening complication of red blood cell (RBC) transfusion. In a prior study, we reported on a total of 35 pediatric patients who had hyperkalemia during or within 12 hours after the RBC transfusion from four facilities in a one-year period (9/1/15 to 8/31/16).¹ This prior study showed the overall prevalence of TAH in the pediatric population is 0.93%, 1-day mortality after TAH is as high as 20%, all patients who experienced TAH were severely ill with multiple comorbidities, median transfused volume within 12 hours before the TAH occurrence was 24% of estimated total blood volume (eTBV), and median infusion rate was high (19.57 ml/Kg/hour). In addition, our research demonstrated that age and irradiation status of RBC units might not be as influential as we expected for the timing of TAH occurrence. Now, we collected data from the patients who received RBC transfusion(s) but did not experience TAH in pediatric population as a control group from the same four facilities in the same time period. Two groups of data, the patients who had TAH (TAH group) and those who did not have TAH (non-TAH group), were compared and multivariate analyses was performed for the factors associated with TAH occurrence.

Study Design

Data collection

The medical records of patients who received RBC transfusion(s) and were younger than 18 years at the time of transfusion in a same 1-year period (9/1/15 to 8/31/16) in the same four facilities (3 children’s hospitals in large academic medical facilities and 1 large academic children’s hospital, all in the United States) as our last report were investigated. At each facility and for each age group, the control group (non-TAH group) included the first 3 eligible patients who did not experience TAH during the RBC transfusion or had normal follow-up potassium (K+) levels within 12 h after the completion of RBC transfusion(s). Age groups were defined as follows; Group 1: ≤28 days old (neonate), Group 2: 29 days to <1 year old, Group 3: 1–5 years old, Group 4: 6–10 years old, and Group 5: 11-<18 years old. Since many case reports in the literature were reported on neonates^2,3, ≤28 days old group was analyzed separately in this study. Therefore, 12 patients in each age group, a total of 60 patients were investigated as a control group (non-TAH group). Hyperkalemia was defined as K+ level above institutional reference range (upper limit of 4.9–5.5 mmol/L depending on the patient’s age and facility), and TAH was defined as K+ increase >0.2mmol/L during or after the transfusion compared to K+ before the transfusion in the same patient. Minor K+ increase of ≤0.2 mmol/L was excluded as an acceptable variable.

Similar to our previous report¹, data collected in this study included patient demographics, diagnosis, medical history and comorbidities, date and time of transfusion and recorded pre-transfusion K+ levels, pre-transfusion creatinine (Cr) level, and 1-day mortality after the RBC transfusion. The storage age, volume of the unit, blood type and modification of the units (leukoreduced, irradiated, washed, supernatant reduced), total volume of RBCs transfused within 12 hours (tV-12), and date and approximate start and end times of transfusion of each unit were also recorded when available. Irradiation of RBC units was performed based on each facility’s protocol; universal irradiation of RBC units for pediatric patients by default in 3 facilities and on request in 1 facility in the study period. The date and time of irradiation was also collected.

Estimated TBV in each patient was calculated based on patient age at the time of transfusion: premature infant 100ml/Kg, <1 month old 85ml/Kg, <4 months old 75 ml/Kg, and ≥4 months old male 70 ml/Kg and female 65 ml/Kg. The RBC infusion rate (IR) was calculated from the volume of RBCs transfused and start/end time of transfusion.

Statistical analyses

After completion of the data collection at the four facilities, the data analyses were performed in aggregate with the previously collected data of patients who had TAH (TAH group). Due to the skewed distributions of many data elements, median measures of centrality are mainly used to assess group-based statistically significant differences. As appropriate and depending on the variable types, the TAH and non-TAH groups were analyzed to assess statistically significant differences quantified by parametric (t-test) and non-parametric (Wilcoxon test, Chi-square test) statistics and their corresponding tail probability values. A default a priori false-positive rate, p-value<0.05, was used to determine statistical significance. In order to perform multivariate analysis on all factors including patients, RBC units and transfusions, only one data was used per patient in all factors for analyses for this study. The corresponding mean value was calculated on IR or age and timing of irradiation of multiple RBC transfusions or for 2 TAH occurrences in the same patient (2 patients had 2 TAH occurrences).

The two-sample Wilcoxon rank-sum test, also known as Mann-Whitney U test, is a nonparametric statistical test that is applicable for comparing the distributions of two populations, or samples^4,5,6. For each outcome of interest – diagnosis (patients vs. controls) and mortality (1-day survival vs. death) – and each observed feature, the Wilcoxon test was used to compare the pairs of distributions, compute the U statistics, and obtain the corresponding p-value. For categorical factor variables, the Chi-square test was used to quantify statistical significance. False discovery rate (FDR) correction is used to adjust the estimated probability values for each test to account for multiple comparisons⁷. Without such FDR correction, the effective false-positive rate will naturally increase with the number of performed statistical tests.

Machine learning (ML) methods for dimensionality reduction⁸ were used to project the high-dimensional observed data into a lower dimensional space (2D or 3D) using linear approaches (e.g., principal component analysis, PCA⁹) and non-linear techniques (e.g., t-distributed stochastic neighbor embedding, t-SNE¹⁰, and uniform manifold approximation and projection, UMAP¹¹).

Recursive feature elimination (REF) ¹² and regularized linear modeling (least absolute shrinkage and selection operator, LASSO^8,13) were employed to identify the most salient features associated with predicting specific clinical outcomes (mortality or diagnosis). Compared to classical models, the LASSO estimates are expected to be closer to the true effects (which are unknown). The estimated effects (for all (un)regularized linear models) correspond to estimated square errors that can be decomposed into independent bias and precision (variance) terms. In regularized models, reduction of the model complexity is associated with coefficient shrinkage.

Random forest (RF) importance was chosen as the strategy for ranking features during recursive feature elimination (RFE). This was driven by prior reports suggesting that RF importance yields high R-squared (good model fidelity), the RF modeling scheme fits the target instead of the residuals, and because bagging models (such as RF) often have better performance compared to boosting models.^{14, 15, 16}

Full logistic regression models were built using the generalized linear modeling (GLM) framework¹⁷. The statistical computing environment R¹⁸ was used to develop a reproducible, end-to-end computational protocol (R-markdown electronic notebook^8,19), which includes the entire protocol, from data import, all preprocessing steps, visualizations, analytics, and results reporting.

Results

The age distributions of 35 TAH group patients and 60 non-TAH group patients are shown in Table 1, and demographics, comorbidities, some conditions of transfusions, and mortality in Table 2. In the non-TAH group, blood loss/iron deficiency anemia was the biggest comorbidities (60% of the patients) followed by respiratory failure (41.7%) and prematurity (23.3%). Statistically significant difference between TAH and non-TAH groups was observed for comorbidities of kidney and liver dysfunctions and anemia, types of access (central line vs. peripheral veins), and 1-day mortality.

Table 1:

Number of the patients by patient age in four facilities

TAH and Non-TAH
Patient age (years old)	≤ 28 days	29 days - 1	1 – 5	6 – 10	11 – <18	Total
TAH group	10 (28.6%)	5 (14.3%)	12 (34.3%)	1 (2.9%)	7 (20.0%)	35 (100%)
Non-TAH group	12 (20%)	12 (20%)	12 (20%)	12 (20%)	12 (20%)	60 (100%)
Total	22	17	24	13	19	95

Table 2:

Demographics, comorbidities, and features of transfusion in TAH and Non-TAH groups

			TAH Group, N (%)	Non-TAH group, N (%)	p
Gender	Female		20 (57.1)	23 (38.3)	0.1180
	Male		15 (42.9)	37 (61.7)
Race	Caucasian		13 (37.1)	20 (3.3)	0.5460
	African American		6 (17.1)	17 (28.3)
	Hispanic		1 (2.9)	5 (8.3)
	Asian		1(2.9)	0 (0.0)
	Other/non-Hispanic/Unknown		14 (40.0)	18 (30.0)
Median	Age		466 days	929 days	0.4760
	Weight		9.8 kg	12.1 kg	0.7386
	Estimated total blood volume		637 mL	826 mL	-
Comorbidities	Prematurity		10 (28.6)	14 (23.3)	0.7474
	Cardiac dysfunction		10 (28.6)	13 (21.7)	1.000
	Respiratory dysfunction		10 (28.6)	25 (41.7)	0.2910
	Kidney dysfunction		17 (48.6)	7 (11.7)	0.0002
	Liver dysfunction		8 (22.9)	4 (6.7)	0.0217
	Intracranial disorders/surgery		4 (11.4)	5 (8.3)	0.3360
	Orthopedic surgery		1 (2.9)	3 (5.0)	1.0000
	Blood loss/Iron deficiency anemia		10 (28.6)	36 (60.0)	0.0274
	Hemoglobinopathy		1 (2.9)	8 (13.3)	0.1872
	Sepsis		4 (11.4)	3 (5.0)	0.2344
	Dialysis		3 (8.6)	3 (5.0)	0.8002
Priming of life support when RBC was transfused		ECMO	6 (17.1)	4 (6.7)	0.2082
		CPB	2 (5.7)	2 (3.3)	0.9778
Types of Access		Central line	7 (20.0)	29 (48.3)	0.0014
		Peripheral veins	8 (22.9)	18 (30.0)
		Unknown	20 (57.1)	13 (21.7)	-
Mortality after hyperkalemia occurrence or transfusion		In 1 day	7 (20.0)	0 (0.0)	0.0012
		In 2–7 days	4 (11.4)	2 (3.3)	-
		Within 7 days	11 (31.4)	2 (3.3)	-

ECMO: extracorporeal membrane oxygenation, CPB: cardiopulmonary bypass

(%): ratio of subject number on total patient number in each group

Bold numbers show frequency of ≥20%

The data for patient’s body weight, pre-transfusion Cr and K+, age of the units transfused, tV-12 per weight (ml/Kg), tV-12 per eTBV (ratio), IR, IR per weight (ml/Kg/hour), IR per eTBV (ratio / hour) by age groups are shown in TAH and non-TAH groups separately, in Table 3. Comparing the two cohorts, pre-transfusion Cr, tV-12/Kg, and tV-12/eTBV showed statistical group differences (Table 4: p=0.0118 for Cr, p=0.0229 for tV-12/Kg, and p=0.0436 for tV-12/eTBV).

Table 3:

Patients and RBC units on transfusions

TAH Group		Age groups of tde patients					Median (Total N)
		≤ 28 d	29 d – < 1 yo	1–5 yo	6–10 yo	11–<18 yo
Patients (N or median)	Patients, N (% of total)	10 (28.6)	5 (14.3)	12 (34.3)	1 (2.9)	7 (20.0)	(35)
	Weight (kg)	2.97 (0.68–5.06)	3.03 (1.0–5.3)	11.70 (6.7–21.7)	28.00 (28)	48.46 (25–66.6)	9.80
	Pre-transfusion Cr (mg/dL)	1.07 (0.5–2.1)	0.40 (0.16–1.6)	0.39 (0.2–2.3)	1.50 (1.5)	0.70 (0.2–6.9)	0.70
	Pre-transfusion K+ (mmol/L)	5.6 (3.4–6.8)	4.9 (3.6–6.6)	4.2 (3.1–5.2)	4.0 (4.0)	4.3 (3.6–5.1)	4.4
	Hyperkalemia occurrences (N)	11	6	12	1	7	(37)
RBC units and Transfusion (N or median)	Units transfused (N)	17	8	29	1	24	(79)
	Units with transfusion time available (N)	15	6	22	1	19	(63)
	Age of units (days)	5.5 (3–15)	5.0 (5–15)	9.1 (5–24)	6.0 (6)	11.4 (9–25)	8.0
	tV-12 (ml)	51.0 (22–505)	57.5 (24–650)	263.0 (60–2950)	280.0 (280)	750.0 (280–2729)	250.0
	tV-12 / Kg (ml/Kg)	43.6 (9.9–123.2)	23.3 (12.3–252.9)	17.0 (5.3–283.7)	10.0 (10.0)	15.5 (4.4–62.6)	23.2
	tV-12 / eTBV	0.48 (0.11–1.52)	0.23 (0.12–3.89)	0.25 (0.08–4.36)	0.15 (0.15)	0.24 (0.07–0.98)	0.26
	IR (ml/hour)	37.8 (7.3–398.0)	67.7 (3.3–487.5)	51.0 (21.6–2011.4)	227.0 (227.03)	112.0 (80.3–1171.9)	76.6
	IR/Kg (ml/Kg/hour)	13.2 (3.0–123.2)	5.0 (3.2–189.7)	4.4 (1.9–193.4)	8.1 (8.11)	7.1 (1.8–33.6)	6.9
	IR/eTBV (/hour)	0.16 (0.04–0.32)	0.04 (0.03–2.92)	0.04 (0.03–2.98)	0.12 (0.12)	0.10 (0.03–0.52)	0.07
Non-TAH Group		Age groups of the patients					Median (Total N)
		≤ 28 d	29 d – < 1 yo	1–5 yo	6–10 yo	11–<18 yo
Patients (N or median)	Patients. N (% of total)	12 (20.0)	12 (20.0)	12 (20.0)	12 (20.0)	12 (20.0)	(60)
	Weight (kg)	2.94 (0.6–4.7)	3.34 (2.3–7.6)	12.10 (8.4–20.3)	26.85 (18.1–41.8)	49.25 (27.6–61.2)	12.10
	Pre-transfusion Cr (mg/dL)	0.58 (0.28–0.9)	0.23 (0.1–0.77)	0.30 (0.19–4.8)	0.40 (0.25–2.3)	0.47 (0.3–0.7)	0.40
	Pre-transfusion K+ (mmol/L)	4.5 (3.3–5.4)	3.8 (3.0–5.3)	4.3 (3.1–4.9)	3.8 (2.8–4.8)	4.2 (3.4–5.4)	4.2
	Hyperkalemia occurrences (N)	0	0	0	0	0	(0)
RBC units and Transfusion (N or median)	Units transfused (N)	14	12	12	15	16	(69)
	Units with transfusion time available (N)	14	12	12	13	15	(66)
	Age of units (days)	8.0 (4–14)	9.0 (4–31)	16.0 (6–42)	18.5 (2–41)	18.0 (1–33)	13.5
	tV-12 (ml)	59.0 (21–82)	68.5 (48–140)	212.5 (87–350)	297.0 (150–520)	434.5 (145–700)	179.5
	tV-12 / Kg (ml/Kg)	19.1 (4.8–33.3)	16.9 (12.8–22.1)	15.2 (10.0–30.8)	11.0 (4.9–20.2)	8.5 (4.4–19.0)	14.1
	tV-12 / eTBV	0.21 (0.06–0.33)	0.22 (0.15–0.30)	0.23 (0.14–0.44)	0.16 (0.08–0.31)	0.13 (0.07–0.27)	0.20
	IR (ml/hour)	18.7 (8.4–62.3)	22.0 (16.2–300.0)	81.1 (38.7–286.6)	110.8 (58.7–391.0)	172.0 (80.9–331.6)	73.0
	IR/Kg (ml/Kg/hour)	8.5 (3.0–20.9)	6.6 (3.2–120.5)	5.9 (4.1–27.6)	4.0 (2.3–20.2)	3.2 (0.04–12.0)	5.3
	IR/eTBV (/hour)	0.10 (0.05–0.21)	0.08 (0.05–1.20)	0.09 (0.06–0.39)	0.06 (0.03–0.31)	0.05 (0.02–0.17)	0.07

p: p-value for difference of Statistics* between TAH and non-TAH group

yo: years old

Cr: creatinine (mg/dL)

K+: potassium (mmol/L)

tV-12: transfused RBC volume within 12 hours before hyperkalemia occurrence

eTBV: estimated total blood volume

IR: infusion rate

Median: the corresponding median of across all patients. To establish homologies across patients and cohorts, only one data is computed for each patient in each category; for patients with observed multiple data points, the mean of those represents the single point estimate.

Table 4:

Between-group differences of all factors used for multivariate analyses for TAH occurrence, assessed by using nonparametric test (chi-square test) * and two-sample Wilcoxon (Mann-Whiney) test, as appropriate

Covariates		p-value
Sex*		0.1180
Race*		0.5460
Access*		0.0012
Age (days)		0.4760
Body weight (Kg)		0.7386
Pre-transfusion K+ level		0.1158
Pre-transfusion Cr		0.0118
Comorbidities	Prematurity	0.7474
	Cardiac dysfunction	1.0000
	Respiratory dysfunction	0.2910
	Kidney dysfunction	0.0002
	Liver dysfunction	0.0217
	Intracranial disorders/surgery	0.3360
	Orthopedic surgery	1.0000
	Blood loss/Iron deficiency anemia	0.0274
	Hemoglobinopathy	0.1872
	Sepsis	0.2344
	Dialysis	0.8002
Priming	ECMO	0.2082
	CPB	0.9778
RBC unit features	RBC unit type	0.3661
	Age of RBC unit	0.1101
	Leukoreduction	0.4616
	Irradiation	0.9322
	Washing	0.7257
	Plasma reduction	1.0000
Transfusion features	tV-12/Kg	0.0229
	tV-12/eTBV	0.0436
	IR/Kg/hour	0.4760
	IR/eTBV/hour	1.0000
	Time from irradiation to transfusion	1.0000

Table 4 shows the results of the quantitative comparison (FDR-adjusted p-values corresponding to the Wilcoxon test-statistics) between TAH and non-TAH groups for TAH occurrence for factors analyzed with subsequent multivariate analyses. We attempted to find and report the corresponding cutoff values (thresholds) that may be used to segregate higher and lower risk patients: the cutoff for tV-12/Kg was 30 ml/Kg, tV-12/eTBV was 0.3 (30% of eTBV), and the age of the RBC unit was 7.95 days for TAH occurrence (Figure 1). The distributions of other variables for TAH occurrence and all variables for 1-day mortality (data available in the link at the end of discussion) are not highly separable, and such cutoffs are not useful therefore not reported.

Figure 1:

Examples of several feature distributions for TAH patients and controls, along with empirical estimates of threshold (cut-off) values suggesting approximate lower and higher risk markers (dotted line)

Figure 1–1: Distributions of total transfused blood volume within 12 hours per Kg

Figure 1–2: Distributions of total transfused blood volume within 12 hours per patient’s estimated total blood volume

Figure 1–3: Distributions of age of RBC units

A similar ratio of RBC units was irradiated in both groups (Table 5). As for the effect of irradiation on TAH occurrence, neither the median age of the RBC units nor the median/mean time from irradiation to transfusion were statistically significant between the two cohorts (Wilcoxon p-value of 0.11). All RBC units are leukoreduced and none of those units are neither washed nor volume reduced in TAH group. Three patients received 1 washed unit each and 1 patient received 1 volume reduced unit in non-THA group. The difference between groups is too small and not meaningful and multivariate analyses also showed no significant differences for all modifications on RBC units.

Table 5:

Comparison of effects of factors that may increase extracellular K+ concentration between TAH and Non-TAH groups

	TAH group (35 patients)	Non-TAH group (60 patients)	P
Irradiated units, N (% of total)	49 (62.0% of 79 units)	44 (63.8% of 69 units)	0.9322
Age of the units (day), median	8.0	13.5	0.1101
Time from irradiation to transfusion (min), median	118.0	118.0	0.9724
Time from irradiation to transfusion (min), mean	1239.7	1421.0

Table 6 shows the coefficients (effect-size estimates) obtained for fitting a regularized LASSO linear model to predict TAH diagnosis. As the LASSO model does model estimation jointly with feature selection, only these variables in the table are included in the model with the corresponding coefficients. Penalized LASSO estimates the same quantities – model coefficients or effects, as classical unpenalized maximum likelihood models. The numerical values of LASSO estimates are generally different from their counterparts generated by unpenalized regression. Many coefficients are annihilated, whereas others are shrunk, i.e., and transformed closer to zero. LASSO tends to annihilate all highly correlated variables and only preserve a single variable as a representative of the general direction of a bundle of highly dependent features. This shows tV-12/Kg is the biggest factor associated with TAH occurrence followed by age of RBC unit, comorbidity of respiratory dysfunction, liver dysfunction, pre-transfusion K+, anemia, tV-12/eTBV and kidney dysfunction. Figure 2 shows a box-and-whisker plot contrasting the variable distributions for non-TAH group (green) and TAH group (blue). In a similar LASSO model for predicting 1-day mortality, none of the variables showed statistically significant group differences (data available in the link at the end of discussion). Figure 3 illustrates one example of the dimensionality reduction approach for rendering all data in 2D and making UMAP predictions of TAH occurrence using the other covariates. In this figure, alphanumerical (1 and 2) labels and color-coded scatter-points represent respectively instances of the actual observed TAH phenotypes (1=non-TAH, 2=TAH) and their corresponding UMAP-predicted counterparts (green and red colors). The close clustering of the UMAP predictions, matching the observed data phenotypes (non-TAH, TAH), suggests that non-linear dimensionality reduction (UMAP) applied to the entire dataset captures the intricate relations and complex associations in this high-dimensional clinical dataset. Additional t-SNE and UMAP examples are shown in electronic format on the supporting website (see below).

Table 6:

Results of multivariate analysis (LASSO regularized linear modeling) for TAH prediction

	Variable	Effect
	(Intercept)	−2.6648
1	tV-12/Kg	0.0099
2	Age of RBC unit	−0.0158
3	Respiratory dysfunction	−0.0940
4	Liver dysfunction	0.1421
5	Pre-transfusion K+	0.1586
6	Anemia	−0.3909
7	tV-12/eTBV	0.8243
8	Kidney dysfunction	1.3068

Only the effect-sizes (linear model coefficients) of the significant features reported by the LASSO model are shown. In principle, the estimates of the standard errors of the effects can be calculated using the bootstrap method. However, these are not reported as in regularized linear models the standard errors are not particularly interpretable. Penalized estimation reduces the variance of estimators by compromising estimate bias (cf. bias-precision tradeoff). The bias of each estimator represents a major component of its mean squared error, and its variance may contribute very little to the MSE.

Figure 2:

The feature distributions of non-TAH (green) and TAH (blue) groups for TAH occurrence. Some of the graph is truncated to show the rough distributions of the bulk of the variables in linear scale along the vertical axis.

Figure 3:

Example of UMAP dimensionality reduction

Discussion

Our previous report¹ showed pediatric patients of all ages could potentially experience TAH and those patients had high 1-day mortality (20%) after the hyperkalemia occurrence. Now, we investigated pediatric patients who received RBC transfusion(s) but did not develop TAH as a control group and analyzed the differences between TAH group and non-TAH group. We also investigated risk factors on TAH occurrence using multivariate analyses.

When demographics and comorbidities in 2 groups are compared (Table 2), the difference with statistical significance was found in kidney and liver dysfunctions, anemia, types of access (central line vs. peripheral veins), and 1-day mortality. TAH group shows higher patients’ ratio in many comorbidities, which may represent that the patients in TAH group are overall sicker than patients in non-TAH group. This fact might contribute to statistically higher 1-day mortality in TAH group in addition to TAH as a possible cause of mortality. Noticeably, none of the patients in the non-TAH group died within 1-day after the RBC transfusion.

Both extracorporeal membrane oxygenation (ECMO) and cardiopulmonary bypass (CPB) require a high volume of RBCs for priming. However, both of these factors did not show the difference between 2 groups and were also not risk factors in multivariate analysis. Types of access was statistically different between two groups when only central line access and peripheral vein access were compared. However, these results may not be statistically valuable because the number of the patients with unknown access type is more than half in TAH group. Peripheral intravenous access for RBC transfusion was recommended by pediatric anesthesiology to avoid cardiac arrest associated with TAH. ² However, our study was conducted for the patients who had not only cardiac arrest associated with TAH but also TAH without cardiac arrest, therefore, the recommendation for access to prevent TAH cannot be made from our data.

When the pre-transfusion Cr, RBC units and transfusion features were compared, only pre-transfusion Cr, tV-12/Kg, and tV-12/eTBV showed statistically significant differences (Table 4). Median of pre-transfusion Cr was 0.7 and 0.4 mg/dL in TAH and non-THA group, respectively. Considering the lower reference range of Cr (0.2–0.6 mg/dL) for patients ≤4 years old (1,460 days old), TAH group has slightly higher median Cr level than this reference range. Therefore, even slightly elevated pre-transfusion Cr may be something that we should pay attention to in addition to history of kidney dysfunction.

Our previous report showed TAH occurred with only 24% of patient’s eTBV within 12 hours. Our recalculated transfused volume with mean value of multiple transfused units also shows only 26% of eTBV. We thought transfusion volume may not be as influential as we expected from this data, however, our multivariate analysis states that the transfusion volume is the most important factor with low threshold; our study shows 30 ml/Kg of transfusion and 30% of eTBV are the prediction threshold on TAH occurrence. In contrast, IR was not statistically different between 2 groups and was not a factor associated with TAH occurrence in our multivariate analysis. However, IR was faster than safe transfusion thresholds of 15 ml/Kg/5h (3 ml/Kg/hour) ²⁰ in both groups, and TAH groups shows mildly faster IR than non-TAH group, which may still suggest that IR needs to be paid attention although it is assumed that slowing IR might be difficult for those severely ill patients.

The age of the RBC units was not statistically different between 2 groups with younger median age in the TAH group. This is probably a result of our already implemented effort to give younger units to younger patients. However, our multivariate analysis shows that the age of the RBC units is the second biggest factor for TAH occurrence (Table 6) with the prediction threshold of 7.95 days old (Figure 1–3). Therefore, it may be recommended to provide RBC units younger than 8 days old to pediatric patients with multiple comorbidities especially kidney dysfunction when institutional protocol of 5 days old RBC units are not available. The dichotomy between the relatively low statistical significance and the high practical importance of age of the RBC units may be explained by the non-parametric distribution of the data.

Table 5 shows the effects of irradiation. It is well known that irradiation increases extracellular K+ in RBC units. However, our previous report demonstrated that irradiation did not have much effect on hyperkalemia occurrence in the groups that had hyperkalemia during transfusion vs. after transfusion. Interestingly, our current analyses also demonstrated that the irradiation did not cause the differences on TAH occurrence in the 2 groups with seemingly similar status of extracellular K+ levels in RBC units by irradiation status. This may be because the facilities utilized protocols to provide freshly irradiated RBCs when available, but it may suggest the pre-transfusion extracellular K+ level in RBC unit may not be as influential as we have believed. K+ depleted RBCs in RBC unit is thought to absorb K+ once they are transfused into the body, and in fact, hypokalemia is reported to be associated with massive transfusions in adult patients.^21,22 Our multivariate analysis showed that age of RBC unit is the second biggest factor on TAH occurrence. These results may imply that important factor to cause TAH may not be an extracellular K+ levels in RBC units per se, rather, other changes of RBCs over the storage time, such as decreased ATP which takes a role in sodium potassium pump and the complex metabolic control of K+ on RBCs, may be more important factor. ^23,24 When the sodium potassium pump does not work, influx of K+ can be minimal and extracellular K+ stays in extracellular space. In conjunction with our analyses, current method to create pathogen reduced RBCs should eliminate irradiation requirement to prevent transfusion associated graft-versus- host-disease in theory, therefore, the concern to cause TAH from higher K+ levels in RBC units by irradiation may be eliminated in near future. However, extracellular K+ levels in RBC units may still need to be considered for the high-risk patients with multiple comorbidities, especially kidney dysfunction, for TAH occurrence when only older RBC units with low ATP levels are available for transfusion.

Lastly and importantly, our univariate analysis methods showed the pre-procedure Cr, comorbidity of kidney dysfunction, tV-12/Kg, and tV-12/eTBV as important factors for TAH occurrence (Table 5). LASSO multivariate analysis showed tV-12/Kg and tV-12/eTBV, age of RBC unit, pre-transfusion K+, and kidney dysfunction are the factors associated with TAH occurrence (Table 6). This data confirms kidney dysfunction is an important factor associated with TAH occurrence as we expected. Also, higher risk lies on the RBC age of higher than 7.95 days old, therefore, younger units should still be provided for the pediatric patients who have multiple comorbidities. And we here propose rules of 30, the pediatric patients who need RBC transfusion more than 30 ml/Kg or 30% of eTBV are at higher risk to have TAH occurrence in those severely ill patients.

In addition to the limitations that we reported in our previous study for TAH group such as possible inaccurate transfusion times and difficulties to evaluate the cause of TAH in multiple comorbidities that most patients had, the analyses of the group differences in the current study may be limited by the small number of patients in each age group. For many clinical variables, the relatively large dispersion of the observations limited the precision of the corresponding parameter estimates. These increased the uncertainty in identifying specific thresholds segregating higher and lower risk factors.

As firm supporters of rigorous, transparent, and reproducible science, the entire data analytic protocol is openly shared as an end-to-end R-markdown electronic notebook (https://socr.umich.edu/docs/uploads/2022/Hyperkalemia_Study.html). This allows community validation, further extension, independent utilization, application on new case-studies, and collaborative improvements of the analytical strategy used in this manuscript.

Conclusions

In this study, we aimed to identify factors associated with TAH occurrence using multivariate analyses and artificial intelligence techniques among characteristics of patients, RBC units and transfusions by comparing TAH and non-TAH groups. In summary, tV-12/Kg, and tV-12/eTBV, age of RBC units and kidney dysfunction are found to be important factors, and a number of comorbidities and IR should also be considered as associations with TAH occurrence. Informative thresholds separating higher and lower risk factors could be reliably calculated for only three factors: tV-12/Kg of 30 ml/Kg, tV-12/eTBV of 30%, and RBC unit age of 7.95 days. From these findings in this study, we suggest the following recommendation for RBC transfusion in pediatric patients to avoid TAH occurrence. For patients who have multiple comorbidities especially kidney dysfunction with higher pre-transfusion Cr or K+ level, RBC units of younger than 8 days old are recommended to be used when institutional protocol of 5 days old RBC units are not available. More importantly, clinicians should take caution when the transfusion volume is greater than 30 ml/Kg or more than 30% of eTBV, which could cause TAH in those patients. The authors refer to this clinical recommendation as the “rule of 30”.

The status and timing of irradiation was not found to be associated with TAH occurrence and may not be very important factors in TAH occurrence; therefore, no changes in current clinical practice is recommended at this time. Infusion rate was not a factor associated with TAH occurrence in both univariate and multivariate analyses, but overall IR in both TAH and Non-TAH groups was faster than safe transfusion thresholds. It may be difficult to slow IR and decrease transfusion volume for those severely ill patients, therefore, recognizing risks and preparing treatments may be the foci to prevent TAH and possible subsequent complications.