Prediction of clinically relevant hyperkalemia in patients treated with peptide receptor radionuclide therapy

Background Peptide receptor radionuclide therapy (PRRT) is applied in patients with advanced neuroendocrine tumors. Co-infused amino acids (AA) should prevent nephrotoxicity. The aims of this study were to correlate the incidence of AA-induced hyperkalemia (HK) (≥5.0 mmol/l) and to identify predictors of AA-induced severe HK (>6.0). Methods In 38 patients, standard activity of 177Lu-labelled somatostatin analogs was administered. Pre-therapeutic kidney function was assessed by renal scintigraphy and laboratory tests. For kidney protection, AA was co-infused. Biochemical parameters (potassium, glomerular filtration rate, creatinine, blood urea nitrogen (BUN), sodium, phosphate, chloride, and lactate dehydrogenase (LDH)) were obtained prior to 4 and 24 h after the AA infusion. Incidence of HK (≥5.0) was correlated with pre-therapeutic kidney function and serum parameters. Formulas for the prediction of severe hyperkalemia (>6.0) were computed and prospectively validated. Results At 4 h, HK (≥5.0) was present in 94.7% with severe HK (>6.0) in 36.1%. Values normalized after 24 h in 84.2%. Pre-therapeutic kidney function did not correlate with the incidence of severe HK. Increases in K+ were significantly correlated with decreases in phosphate (r = −0.444, p < 0.005) and increases in BUN (r = 0.313, p = 0.056). A baseline BUN of >28 mg/dl had a sensitivity of 84.6% and a specificity of 60.0% (AUC = 0.75) in predicting severe HK of >6.0 (phosphate, AUC = 0.37). Computing of five standard serum parameters (potassium, BUN, sodium, phosphate, LDH) resulted in a sensitivity of 88.9% and a specificity of 79.3% for the prediction of severe HK >6.0 (accuracy = 81.6%). Conclusions A combination of serum parameters predicted prospectively the occurrence of relevant HK with an accuracy of 81.6% underlining its potential utility for identifying ‘high-risk’ patients prone to PRRT.

For kidney protection, various protocols using co-infusion of positively charged amino acids (AA) like L-arginine and L-lysine have been introduced [6][7][8][9][10]. However, side effects including nausea, vomiting, and clinically relevant hyperkalemia have been described [11,12]. In a recent study including patients with NET undergoing PRRT with 90 Y-DOTATOC, Giovacchini et al. reported the incidence of hyperkalemia (>5.0 mmol/l) in more than three-fourths of patients with potassium levels up to 6.7 mmol/l, thereby raising awareness of this potentially life-threatening adverse effect induced by AAcoinfusion [13].
In clinical routine practice, early or even pre-therapeutic identification of patients at risk for developing severe hyperkalemia (>6.0 mmol/l) due to AA-coinfusion would be of high relevance. Therefore, the aims of this study were to assess the incidence and severity of hyperkalemia (≥5.0 mmol/l) and to identify predictors of PRRT-related severe hyperkalemia (>6.0 mmol/l).

Study design
The study consists of a retrospective and a prospective part. In the retrospective analysis, 38 consecutive patients were included. Incidence and severity of hyperkalemia (≥5.0 mmol/l) were correlated with pre-therapeutic kidney function. Three formulas for the prediction of severe hyperkalemia (>6.0 mmol/l) were computed retrospectively from different combinations of serum parameters. Thereafter, these formulas were prospectively validated in a cohort including another 38 patients.
This study comprised of only retrospective analysis of routinely acquired data and therefore the institutional review board (ethics committee of the Medical Faculty of the University of Würzburg) waived the requirement for additional approval. All patients gave written informed consent to receive standard-of-care PRRT. In October 2013, the routinely administered amount of AA was reduced from 75 g to 50 g to comply with the most recent version of the joint International Atomic Energy Agency (IAEA), European Association of Nuclear Medicine (EANM), and Society of Nuclear Medicine and Molecular Imaging (SNMMI) practical guidance [6].

Patients
From July 2013 to December 2013, 38 consecutive patients (24 males, 14 females) referred for PRRT were enrolled. Their mean age was 62 ± 14 years (range, 24 to 82). The general exclusion criteria, as defined by the joint IAEA, EANM, and SNMMI practical guidance, were applied [6]. All patients had progressive disease. Out of the 38 subjects enrolled, 7 received their first treatment cycle. The following risk factors for the occurrence of kidney toxicity were assessed in all patients: diabetes mellitus (n = 5) and arterial hypertension (n = 12). Additionally, potassium-sparing anti-hypertensive medications and long-term use of NSAIDs (n = 17) were recorded. Pretherapeutic ECG was unremarkable in all patients. None of the subjects had a history of arrhythmogenic diseases.
From January 2014 to April 2014, the formulas for hyperkalemia-risk-assessment derived from the retrospective cohort were validated prospectively in 38 additional consecutive subjects (23 males, 15 females; mean age 58 ± 12 years (range, 24 to 76)). Twenty-two patients were included and monitored during both treatment cycles with co-infusion of 75 g and 50 g of AA, as previously reported [14]. Detailed patient characteristics are given in Table 1.

Assessment of kidney function
About 2 weeks prior to PRRT, 22 patients underwent renal scintigraphy with 99m Tc-labelled mercaptoacetyltriglycine (MAG3) on a single head gamma camera (Signature; Siemens, Erlangen, Germany) equipped with low-energy, high-resolution collimators. The radiotracer was prepared using a commercially available kit (MAG3, Mallinckrodt Pharmaceuticals, Neustadt an der Donau, Germany). All patients were asked to drink at least 250 ml of water 30 min prior to the examination. Imaging started immediately after injection of 98 ± 6 MBq 99m Tc-MAG-3 and was performed according to national guidelines [15]. Venous blood samples were drawn from the arm contralateral to the injection site 20 and 30 min after tracer injection to allow for calculation of tubular extraction rates (TER). Sixteen patients underwent renal assessment by 131 I-hippurate on an outpatient basis and were excluded from this part of the study.

Preparation of 177 Lu-DOTATATE/-TOC and PRRT
177 Lu-DOTATATE/-TOC was prepared with minor modifications as previously reported [14,16]. All patients were admitted 1 day prior to therapy to guarantee adequate hydration (1 l saline) and were hospitalized for a total of 3 days. In the retrospective group, 37.5 g L-arginine hydrochloride and 37.5 g L-lysine hydrochloride (75 g total amount of AA, pH 7.0) were intravenously administered over 4 h beginning 30 min to 1 h prior to PRRT. In the prospective cohort, amounts of AA infused were changed to 25 g of each L-arginine and L-lysine (pH 7.0; single-day 50 g protection-protocol) in order to comply with the joint IAEA, EANM, and SNMMI practical guidance [6]. No other fluids were administered. As recently published, the two different amounts of AA do not result in any difference in incidence or severity of hyperkalemia [14].
Pre-and post-therapeutic blood samples

Analysis and statistics
Student's t test was used to test for simple differences between patients with or without hyperkalemia in various blood parameters. A p value of 0.05 or less was considered to be significant. Moreover, we have assumed that some blood parameters promote and some prevent hyperkalemia. Therefore, it may be possible that the human body tries to restore a homeostatic imbalance by increasing one blood parameter or decreasing another. Such relationships could not be accounted for by Student's t tests. However, linear discriminant analysis makes it possible to classify patients with different conditions (e.g., hyperkalemia) on the basis of a set of predictors (e.g., multiple blood parameters). Hence, we performed multiple linear discriminant analysis to improve the prediction of severe hyperkalemia (>6.0 mmol/l) by calculating linear combinations of the relevant blood parameters.
In a backward stepwise analysis, the weakest serum predictor contributing least to the prediction of severe hyperkalemia was excluded according to its p value. Cutoff values for the prediction of severe hyperkalemia were determined for each formula by receiver operating characteristic (ROC) analysis [17]. Additionally, in order to test a correlation between increases of potassium and clinical factors, Pearson's correlation was performed. A p value of 0.05 or less was considered to be significant.
Five of the 13 subjects (38.5%, 2/5 with diabetes) with severe hyperkalemia became symptomatic with palpitations and general malaise. In all five patients, ECG revealed flattened P waves and high peaked T waves with peak serum potassium levels of 8.5 mmol/ l. These patients required immediate therapy to lower serum potassium concentrations. We effectively administered 10 units of regular insulin in 500 ml of 10% dextrose over 60 min.

Correlation of hyperkalemia to kidney function and clinical factors (retrospective cohort)
All patients presented with normal kidney function according to 99m Tc-MAG3 scintigraphy with a mean TER of 231 ± 47 ml/min/1.73 m 2 . The pre-therapeutic TER did not correlate with the post-therapeutic potassium level (r = −0.120) and was unable to predict incidence or severity of hyperkalemia. No cutoff value of minimum glomerular filtration rate excluding severe hyperkalemia could be derived from serum blood samples (r = −0.055, p = 0.74).
Additional serum parameters (potassium, BUN, sodium, phosphate, chloride, LDH) were analyzed ( Table 2). Out of all variables investigated, a pre-therapeutic BUN cutoff value of >28 mg/dl (defined by ROC analysis) had a sensitivity of 84.6% and a specificity of 60% in predicting severe hyperkalemia ≥6.0 mmol/l (area under the curve, AUC = 0.75). No clinical characteristic including patient age, cumulative dose, number of treatment cycles, potentially interfering medications, or clinical risk factors correlated with the incidence or severity of post-treatment hyperkalemia (Additional file 1: Table S1).  We also investigated changes to the baseline to 4-h post-therapy initiation. Changes in serum BUN tended to positively correlate with hyperkalemia (r = 0.313; p < 0.056) (Figure 2A). A significant negative correlation between increase in serum potassium and phosphate was observed (r = −0.444; p < 0.005). However, this parameter did not predict severe hyperkalemia (>6.0 mmol/l) (AUC = 0.37) ( Figure 2B).

Calculation of formulas for hyperkalemia-risk-assessment (retrospective cohort)
In order to improve estimation of the risk of severe hyperkalemia, several formulas were generated based on an arbitrarily set cutoff potassium level >6.0 mmol/l and different serum parameters. Three of these formulas showed superior results in retrospective evaluation.
GFR and creatinine were omitted in Formula 2: Formula 2 (n = 36): Based on an AUC of 0.90 for Formula 2, the optimal cutoff value was defined as X ≥ −29.226 with a sensitivity of 92.3% and a specificity of 82.6% (accuracy = 86.1%).
Excluding further two parameters (BUN, LDH) while introducing creatinine led to Formula 3:  (accuracy = 78.9%). The respective ROC curves are given in Figure 3.

Validation of formulas for hyperkalemia-risk-assessment (prospective cohort)
In a prospective setup, additional 38 patients were enrolled to validate the formulas: none of the patients suffered from hyperkalemia prior to PRRT with mean potassium levels of 4.34 ± 0.39 mmol/l. 4 h after PRRT, mean K + level was 5.69 ± 0.61 mmol/l with values ≥5.0 mmol/l in 28/38 subjects (73.6%; 15/28 subjects with history of potassium-sparing anti-hypertensive medications and long-term use of NSAIDs). Nine of the 38 patients (23.7%) experienced severe hyperkalemia >6.0 mmol/l, and three patients became symptomatic with palpitations and ECG changes. After 24 h, K + levels returned to baseline levels (mean 4.53 ± 0.37 mmol/l).
The three formulas were validated using the cutoff values mentioned above. Formula 1 showed a sensitivity of 77.8% and a specificity of 79.3% (accuracy = 78.9%) and Formula 2 had a sensitivity of 88.9% and a specificity of 79.3% (accuracy = 81.6%). Formula 3, including four blood values, could predict severe hyperkalemia >6.0 mmol/l with a sensitivity of 66.7% and a specificity of 82.8% (accuracy = 78.9%).
An overview over the results of the three formulas is given in Table 3 (retrospective and prospective cohort).

Discussion
The high incidence of PRRT-related hyperkalemia in patients treated for NET has been recently reported [13]. The aim of this study was to identify pre-therapeutic parameters for prediction of AA-induced hyperkalemia in patients undergoing PRRT. Combinations of serum parameters were retrospectively identified, and their predictive value was prospectively evaluated. No single serum parameter could reliably predict the occurrence of hyperkalemia. In the prospective validation cohort, a formula including five routine biochemical serum parameters (potassium, BUN, sodium, phosphate, LDH; Formula 2) correctly identified 8 of 9 patients with (sensitivity: 88.9%) and 23 of 29 patients without hyperkalemia >6.0 mmol/l (specificity: 79.3%); resulting in an accuracy of 31 of 38 (81.6%).
Showing the highest sensitivity and a false-negative rate of 11.1%, thus really identifying patients suffering from clinically relevant hyperkalemia, Formula 2 seems to be the most promising. Considering a potentially life-threatening side-effect, this approach might be a quite valuable tool for the clinical physician planning PRRT.
Arginine monohydrochloride and lysine monohydrochloride, both cationic amino acids, are known to cause transient hyperkalemia by driving potassium from the intracellular to the extracellular compartment   [18,19]. In our cohort, hyperkalemia (>5.0 mmol/l) was induced in almost every patient and K + levels rose by 33.8%. This finding is in line with previous results [11,13,20]. Thus, close monitoring of patients after PRRT and evaluation of serum potassium levels should be mandatory as more than 35% of subjects (13/36) experienced rises in K + to > 6.0 mmol/l. Five patients became symptomatic and required potassium-lowering interventions (e.g. insulin with glucose). Of note, two of those patients suffered from diabetes (40% of all diabetic patients in this cohort), a sub-cohort maybe especially prone to post-therapeutic hyperkalemia due to reduced sensitivity to endogenous insulin. An earlier study has already reported on this pathophysiologic mechanism [21]. In our study, no significant correlation between diabetes and incidence or severity of increases in potassium could be detected; however, this might be related to small sample sizes.
To prevent potential complications, we aimed to identify predictors for severe hyperkalemia >6.0 mmol/l to help the clinician in identifying patients at risk. The only factor with limited predictive value was an increased serum BUN. A cutoff value of >28 mg/dl provided a sensitivity of 84.6% but a rather low specificity of 60.0% in excluding severe hyperkalemia >6.0 mmol/l (AUC = 0.75). However, since emphasis should be put on high sensitivity, this marker is suggested as simple first test in evaluating potentially severe hyperkalemia.
In a more sophisticated approach, a combination of biochemical markers which are all affected by PRRT was used to derive three different formulas for predicting the risk of AA-induced hyperkalemia. The four-to-seven parameters considered by these formulas can be obtained from one blood sample. All formulas showed excellent areas under the curve in retrospective ROC analysis (range, 0.90 to 0.92).
In order to allow broad applicability, accuracy was prospectively investigated. In this cohort, the formula taking into account five serum parameters (potassium, creatinine, sodium, LDH and phosphate; Formula 2) demonstrated again both excellent sensitivity (88.9%) and good specificity (79.3%).
The following limitations have to be considered: in order to strengthen the statistical power and the prognostic significance of our findings, further studies with larger patient cohorts need to be performed. Secondly, almost more than half of the patients were included and monitored during both retrospective and prospective treatment cycles with co-infusion of 75 or 50 g of AA. Moreover, the amount of AA routinely administered was reduced from 75 to 50 g to comply with the latest joint IAEA, EANM, and SNMMI practical guidance in October 2013 [6]. Thus, the retrospectively calculated formulas were derived administering 75 g AA whereas in the prospective part these findings were validated administering 50 g AA. However, in our intra-individual comparison of both protocols, incidence and severity of hyperkalemia do not significantly vary between 75 and 50 g AA, respectively [14]. Due to this finding of our previous study, a broad applicability in clinical routine is independent from AA dosing (75 or 50 g, respectively). On a final note, different radiolabelled compounds were used. In the prospective cohort, DOTATATE was replaced by DOTATOC. Esser and colleagues reported that therapy with 177 Lu-octreotate is preferable due to its higher tumor residence time. However, this research group did not report on incidence of hyperkalemia after administering both protocols [22].
In conclusion, the high incidence of hyperkalemia in almost all patients undergoing PRRT requires a high degree of clinical attention. Serum blood sampling after treatment is mandatory, given that more than 20% of patients experience a potentially life-threatening rise in K + values to >6.0 mmol/l. Early identification of subjects at risk who might require immediate therapeutic intervention would be of great value. To the best of our knowledge, this is the first attempt to derive a robust estimate of the risk of AA-induced hyperkalemia in patients scheduled for PRRT. With an accuracy of 81.6% and a low false-negative rate, we established a reliable prospectively validated formula assessed by one simple blood collection selecting 'high risk' patients being vulnerable for severe hyperkalemia. In general, this formula might be a valuable tool for the clinical physician planning PRRT and was recently implemented into our clinical routine.