Predictive Capability of Metabolic Panels for Post-Operative Atrial Fibrillation in Cardiac Surgery Patients

Abstract

Background

Post-operative atrial fibrillation (POAF) occurs in up to 65% of cardiac surgery patients and is associated with increased risk for stroke and mortality. Electrolyte disturbances in sodium (Na⁺), potassium (K⁺), total calcium (Ca²⁺), chloride (Cl⁻), and magnesium (Mg²⁺) are predisposing factors for POAF, but these imbalances have yet to be used to predict POAF. The purpose of this study is to determine whether the development of POAF can be predicted by blood plasma ionic composition.

Methods

Metabolic panels of patients with no prior history of AF who did (n = 763) and did not develop POAF (n = 2144) after cardiac surgery were obtained from the Carilion Clinic electronic medical record system. We initially evaluated serum Na⁺, K⁺, Ca²⁺, Cl⁻, and Mg²⁺ in the two groups using descriptive statistics via scatter and spaghetti plots, and then with predictive modeling via logistic regression and random forest models.

Results

Neither scatter nor spaghetti plots of electrolyte data revealed a significant difference between those who did and not develop POAF. Two logistic regression models and two random forest models with POAF status as the outcome were generated using the first observation for each electrolyte and the coefficient of the linear regression, which was obtained from a linear fit of the scatter plot. The random forest model using the first observation, had a sensitivity of only 12.2%, but all four models had specificities greater than 97%.

Conclusions:

Neither of the two logistic regression nor two random forest models were able to effectively predict the development of POAF from plasma ionic concentrations, but the random forest models effectively classified patients that would not develop POAF.

Introduction:

Post-operative atrial fibrillation (POAF) is a complication that has been reported to occur in up to 31.9% of patients who undergo coronary artery bypass grafting (CABG) and in up to 63% of patients who undergo combination CABG and valve replacement surgeries, occurring in most patients within 72 hours¹.

The pathogenesis of AF is multifaceted. In particular, the existence of a structural or ionic substrate has been shown to be strong predisposing factor for AF development. Structural substrates are generally due to direct modification of the atrium, which can be secondary to amyloidosis, age-related fibrosis, or tissue remodeling². Examples of ionic substrates include changes to L-type Ca²⁺ channels responsible for contraction and the plateau phase of myocardial action potentials, alterations in K⁺ channels², or direct electrolyte imbalances. Cumulatively, these substrates increase an individual’s risk for AF, but may not be sufficient for acute presentation. Cardiac surgery induces pro-arrhythmic inflammation, oxidative stress, and sympathetic activation of atrial cardiomyocytes, which can precipitate POAF in those with existing substrates².

To minimize the impact of ionic substrates, current standard of care advocates for tight control of electrolytes to minimize risk of POAF. From current literature, among commonly measured serum electrolytes, hyponatremia, hypokalemia, and hypomagnesemia have been identified as particular risk factors. There are an increasing number of studies evaluating the roles of calcium and chloride on AF development, but their exact roles remain unclear. Establishment of these electrolytes as independent predictors may allow clinicians to identify patients at high risk for AF in the post-operative setting with a simple lab test. While some studies have assessed the utility of other clinical biomarkers in predicting POAF^3,4, none have attempted similar analyses with electrolytes.

Therefore, this study was designed to test the hypothesis that Na⁺, K⁺, Ca²⁺, Cl⁻, and Mg²⁺ obtained from metabolic panels can be used to predict patients at high risk for the development of post-operative atrial fibrillation.

Methods:

Study Population

The study was approved by the Institutional Review Board of the Carilion Clinic health system in Roanoke, VA, USA (protocol #2406). Using The Society of Thoracic Surgeons database, 3492 operative cardiac procedures performed on patients 18 years of age or older at Carilion Clinic between August 1, 2008 and September 30, 2017 were cross-referenced with their respective encounter in the electronic medical record. An IRB approved waiver was obtained for patient consent. Data extracted included age, sex, prior history of atrial fibrillation, date and type of procedure, procedural times, hospital length of stay, the ionic components of the first ten basic metabolic panels (BMP) plus calcium and magnesium drawn after the end of the procedure, whether the patient experienced POAF, and the time of POAF if applicable. Atrial fibrillation was defined according to the Heart Rhythm Society definition as “any arrhythmia that has the ECG characteristics of AF and lasts sufficiently long for a 12-lead ECG to be recorded, or at least 30 seconds on a rhythm strip”⁴. Additionally, further descriptive variables including patient’s BMI, smoking status, history of hypertension/diabetes/chronic lung disease/myocardial infarction/heart failure and NYHA class/dialysis, and whether they were taking ACE inhibitors, beta blockers, or statins at the time of admission were obtained. 518 patients with a pre-existing history of AF were excluded from the study. After exclusion criteria were applied, of the remaining patients, 2207 patients did not develop AF (NoAF) and 767, or 25.79% of the patients, developed POAF. A further 3 patients were excluded from each group due to missing or improperly recorded electrolyte data.

Statistical Analysis

Descriptive statistics were generated to characterize the overall sample of patients, according to their eventual POAF status (NoAF, POAF). Categorical variables, such as sex and procedure type were quantified by counts and percentages; whereas means, standard deviations, medians, interquartile ranges, and ranges were used to describe continuous variables, including age and the initial electrolyte readings. Every initial electrolyte reading was taken within 36 hours post-operation.

Data was visualized with scatterplots of the initial electrolyte readings, and within-patient changes over time were visualized with spaghetti plots.

Two predictive models were created from the data. The first included the patient’s sex, type of surgery, age, and their initial electrolyte values after surgery for each of the five ions. The second used sex, procedure code, age, and a set of eight regression coefficients. To obtain these coefficients, simple linear models were fit for each patient using each of the five ions as the outcome and time from the procedure as the predictor. Each model returned two coefficients, a slope and an intercept for each of the ions.

Logistic regression and random forest modeling approaches were selected to examine predictors of POAF. A logistic regression was selected for the first set of our predictive models as it is frequently the standard model used to study a dichotomous outcome. Random forests were selected for our second set as they typically have superior performance for this type of data set. Both of these models analyzed the effect of patient’s sex, type of surgery, and age. We then included the ionic values from the first set of labs after surgery in both models and compared them side-by-side with the linear regression coefficient of each ion as a function of time. These models were then evaluated using predictive metrics, such as area under the curve (AUC), sensitivity, and specificity. These metrics were computed in-sample for logistic regression and out-of-bag for the random forests. The random forests were each constructed using 2,000 classification trees. Each tree is a classifier predicting whether each patient developed POAF. The result of the entire forest ensemble is a predicted probability of developing POAF, which is the proportion of trees which classify a given patient as having POAF⁵.

Results:

Descriptive Statistics

Patient demographic information is summarized in table 1 below. Of note, the type of procedure, patient age, duration of cross-clamp time, duration of cardiopulmonary bypass, smoking status, history of hypertension, and whether the patient was on a beta blocker were all statistically significant at an alpha of 0.05. Specifically, operations that involved valvular procedures, advanced age at the time of the procedure, and a history of hypertension were associated with higher risk of POAF. Both increased duration cross-clamp time and cardiopulmonary bypass time also correlated with higher risk of POAF. Unexpectedly, the number of active smokers was significantly reduced in the POAF patient group. Lastly and as anticipated, patients who had contraindications against beta blocker use developed POAF more often.

Table 1:

Descriptive statistics comparing covariates by Afib status

	Overall Sample (N=2974)	POAF (N=767)	NoAF (N=2207)	P-value
SEX				0.2982
Female	964 (32.4%)	237 (30.90%)	727 (32.94%)
Male	2010 (67.6%)	530 (69.10%)	1480 (67.06%)
PROCEDURE CATEGORY				<0.001
CABG	1672 (56.2%)	377 (49.15%)	1295 (58.68%)
CABG + Valve	168 (5.6%)	54 (7.04%)	114 (5.17%)
Other	499 (16.8%)	132 (17.21%)	367 (16.63%)
Valve	635 (21.4%)	204 (26.60%)	431 (19.53%)
AGE				<0.001
n	2974	767	2207
Mean	64.52	68.38	63.17
SD	12.33	10.28	12.70
Median	66.00	69.00	64.00
Q1, Q3	57.00, 73.00	62.00, 75.00	56.00, 72.00
Min, Max	17.00, 98.00	28.00, 98.00	17.00, 93.00
BMI				0.623
n	2973	767	2206
Mean	29.45	29.36	29.48
SD	5.81	5.96	5.75
Median	28.70	28.70	28.70
Q1, Q3	25.30, 32.90	25.10, 33.10	25.40, 32.80
Min, Max	14.30, 56.00	14.30, 51.00	16.60, 56.00
CROSS CLAMP TIME (MIN)				<0.001
n	2974	767	2207
Mean	79.64	86.90	77.00
SD	44.09	50.4	41.30
Median	70.00	74.50	69.00
Q1, Q3	51.00, 95.00	53.00, 106.00	51.00, 91.00
Min, Max	7.00, 424.00	11.00, 424.00	7.00, 403.00
CARDIOPULMONARY BYPASS TIME (MIN)				<0.001
n	2974	767	2207
Mean	111.19	122.53	107.38
SD	60.32	66.21	57.64
Median	97.00	104.00	93.00
Q1, Q3	73.00, 133.00	77.00, 151.00	72.00, 127.00
Min, Max	17.00, 588.00	21.00, 588.00	17.00, 560.00
SMOKER				0.012
No	2272 (76.4%)	616 (80.31%)	1656 (75.03%)
Unknown	8 (0.3%)	2 (0.26%)	6 (0.27%)
Yes	694 (23.3%)	149 (19.43%)	545 (24.69%)
HYPERTENSION				0.0015
No	561 (18.9%)	112 (14.60%)	449 (20.34%)
Unknown	2 (0.1%)	0	2 (0.09%)
Yes	2411 (81.1%)	655 (85.40%)	1756 (79.57%)
DIABETES				0.2572
No	1787 (60.1%)	479 (62.45%)	1308 (59.27%)
Unknown	1 (0.0%)	0	1 (0.05%)
Yes	1186 (39.9%)	288 (37.55%)	898 (40.69%)
CHRONIC LUNG DISEASE				0.0901
Mild	292 (9.8%)	82 (10.69%)	210 (9.52%)
Moderate	157 (5.3%)	49 (6.39%)	108 (4.89%)
No	2255 (75.8%)	557 (72.62%)	1698 (76.94%)
Severe	89 (3.0%)	28 (3.65%)	61 (2.76%)
Unknown	123 (4.1%)	30 (3.91%)	93 (4.21%)
Unknown Severity	58 (2.0%)	21 (2.74%)	37 (1.68%)
PRIOR MI				0.5068
No	1789 (60.2%)	475 (61.93%)	1314 (59.54%)
Unknown	12 (0.4%)	3 (0.39%)	9 (0.41%)
Yes	1173 (39.4%)	289 (37.68%)	884 (40.05%)
PRIOR HEART FAILURE				0.5966
No	2243 (75.4%)	582 (75.88%)	1661 (75.26%)
Unknown	169 (5.7%)	38 (4.95%)	131 (5.94%)
Yes	562 (18.9%)	147 (19.17%)	415 (18.80%)
DIALYSIS				0.823
No	2925 (98.4%)	754 (98.31%)	2171 (98.37%)
Unknown	1 (0.0%)	0	1 (0.05%)
Ye	48 (1.6%)	13 (1.69%)	35 (1.59%)
ACE INHIBITORS				0.0836
No	2249 (75.6%)	601 (78.36%)	1648 (74.67%)
Unknown	5 (0.2%)	2 (0.26%)	3 (0.14%)
Yes	720 (24.2%)	164 (21.38%)	556 (25.19%)
BETA BLOCKERS				0.0332
Contraindicated	148 (5.0%)	49 (6.39%)	99 (4.49%)
No	245 (8.2%)	52 (6.78%)	193 (8.74%)
Yes	2581 (86.8%)	666 (86.83%)	1915 (86.77%)
STATINS				0.0703
No	1105 (37.2%)	310 (40.42%)	795 (36.02%)
Unknown	8 (0.3%)	1 (0.13%)	7 (0.32%)
Yes	1861 (62.6%)	456 (59.45%)	1405 (63.66%)

Note: P-values are based on a two-sample t-test for continuous variables and chi-square tests for categorical variables.

Initial Na⁺, K⁺, Ca²⁺, Cl⁻ and Mg²⁺ readings in each group by mean, median, standard deviation, their first and third quartiles, and their minimum and maximum values are presented in Table 2. Of note, patients who did not go on to develop POAF had a significantly lower mean Na⁺, 140.68 mEq/L vs. 141.08 mEq/L. There were no statistically significant differences in K⁺, Ca²⁺, Cl⁻, or Mg²⁺.

Table 2:

Descriptive statistics for initial ion readings for Na⁺, K⁺, Ca²⁺. Cl⁻, and Mg²⁺ in the overall sample and by AFib status (N=2970)

	Overall Sample (N=2968)	POAF (N=764)	NoAF (N=2204)	P-value
INITIAL NA READING				0.0185
n	2952	764	2188
Mean	140.79	141.08	140.68
SD	4.05	4.32	3.95
Median	141.00	141.00	141.00
Q1, Q3	138.00, 143.00	138.00, 144.00	138.00, 143.00
Min, Max	123.00, 162.00	123.00, 157.00	124.00, 162.00
INITIAL K READING				0.127
n	2952	764	2188
Mean	4.23	4.26	4.22
SD	0.50	0.52	0.49
Median	4.20	4.20	4.20
Q1, Q3	3.90, 4.50	4.00, 4.50	3.90, 4.50
Min, Max	2.70, 7.60	2.70, 7.40	2.80, 7.60
INITIAL CA READING				0.2694
n	2952	764	2188
Mean	8.65	8.62	8.65
SD	0.71	0.70	0.71
Median	8.70	8.60	8.70
Q1, Q3	8.10, 9.10	8.10, 9.10	8.10, 9.20
Min, Max	5.20, 12.80	5.20, 10.70	5.20, 12.80
INITIAL CL READING				0.2053
n	2952	764	2188
Mean	104.69	104.51	104.75
SD	4.67	4.88	4.60
Median	105.00	105.00	105.00
Q1, Q3	102.00, 108.00	102.00, 108.00	102.00, 108.00
Min, Max	84.00, 121.00	85.00, 121.00	84.00, 120.00
INITIAL MG READING				0.8766
n	2904	760	2144
Mean	2.23	2.22	2.23
SD	0.38	0.37	0.38
Median	2.20	2.20	2.20
Q1, Q3	2.00, 2.40	2.00, 2.50	2.00, 2.40
Min, Max	0.80, 3.70	1.20, 3.70	0.80, 3.70

Note: P-values are based on a one-way ANOVA test for continuous variables.

Descriptive Plots

The ionic concentration plots in Figure 2 are colored by AF status, with blue representing patients that went into POAF and red as those who did not. The horizontal axis is the number of days after the operation the BMP was taken, and the vertical axis is the raw ion value. Each vertical axis is on its own scale.

Figure 2:

Scatterplots of the five electrolyte readings over time

As shown in Figure 2, ionic levels appear to vary with time. There are no statistically significant differences between the values of the POAF measurements and the NoAF measurements

Predictive Models

The NOAF and POAF patient groups were then analyzed using logistic regression models and random forests with AF status as the outcome. Each random forest consisted of 2,000 classification trees. Each tree is a classifier that predicts if a patient develops POAF.

As seen in figure 3, receiver operating characteristic (ROC) curves were generated to better visualize the strength of these models. Areas under the curve (AUC) were calculated for each of the four models, as seen in Table 3. The random forest model on the linear regression coefficients had the largest AUC, though none of the models perform especially well given that even the best AUC was less than 0.75.

Figure 3:

ROC curves for four models.

Table 3:

AUCs for the four models considered in the analysis.

Model	Data	AUC
Logistic	First Observation	0.662
Logistic	Linear Regression Coefficients	0.724
Random Forest	First Observation	0.716
Random Forest	Linear Regression Coefficients	0.641

We then calculated the positive predictive value, sensitivity, negative predictive value, and specificity for the models, as seen in Table 4. Even the best model by this metric, the random forest with the first observation data set, only correctly predicts POAF in 28.6% of cases. However, all the models share a high specificity and negative predictive value, suggesting potential utility for electrolyte panels in the use of determining patients at low risk for POAF.

Table 4:

Predictive model diagnostics for the four models

Model	Data	PPV	Sensitivity	NPV	Specificity
Logistic	First Observation	0.368	0.034	0.747	0.980
Logistic	Linear Regression Coefficients	0.480	0.067	0.751	0.975
Random Forest	First Observation	0.446	0.179	0.767	0.930
Random Forest	Linear Regression Coefficients	0.586	0.286	0.789	0.930

Discussion

POAF is a major cause of morbidity and mortality in patients who undergo cardiac surgery, occurring in nearly one-third of patients who undergo CABG, and two-thirds of patients who undergo combination CABG and valve replacement surgery. Most episodes are self-limiting but can increase risk of post-operative stroke. In a study of over 340,000 patients admitted to 375 US hospitals for surgery, POAF increased mortality from 2.1% to 14.1%³. Even in uncomplicated cases, POAF diverts medical and nursing resources and can significantly prolong hospital stay. Specifically, the median length of stay for patients without POAF was 4 days and 11 days for those who experienced POAF, with the average cost of hospitalization being $15,076 for patients without POAF and $27,164 for patients with POAF⁶.

The majority of patients who experience POAF with no prior history of atrial fibrillation (AF) do so only temporarily, with 15-30% of cases reverting to sinus rhythm within 11-12 hours and as many as 80% reverting within 24 hours⁷. However, in some patient subgroups, POAF can progress to persistent AF, especially in those with underlying risk factors. These patients require lifelong anticoagulation and rate/rhythm control. Prognosis is further worsened in those with co-existing cardiovascular disease, which frequently includes patients undergoing cardiac surgery⁸.The purpose of this study was to determine whether an electrolyte signature can identify patients at high risk for the development of post-operative atrial fibrillation. Specifically, this retrospective chart review study examined ionic substrates by the use of post-operative metabolic panels from patients who underwent cardiac surgery. We found mean serum Na⁺ immediately after surgery to be higher in patients who went on to develop POAF, but no significant difference in any of the other ions. Moreover, we found serum Na⁺, K⁺, and Cl⁻ decreased over the course of hospitalization and Ca²⁺ increased, but no statistically significant differences in the rate of change between the two patient groups. Mg²⁺ remained relatively constant. Using electrolyte data, we were unable to effectively predict the development of POAF using logistic regression or random forest predictive modeling.

Prior studies have evaluated individual ionic thresholds for determining the incidence of POAF. In contrast to the results of our study, prior research has demonstrated some association of hyponatremia with POAF⁹. The exact mechanism is unclear, but the influence of sodium on the sympathetic nervous system via activation of the renin-angiotensin system is thought to play a role. In one study by Cavosuglu et al., POAF was present in 33.3% of patients with hyponatremia and 18.8% of patients with normal Na⁺. Serum sodium of less than 135 mmol/L was able to predict POAF with a sensitivity of 36% and specificity of 79.2%. When that cutoff was set to less than 130 mmol/L, the specificity increased to 95%⁹. While our data indicate that sodium is higher in POAF patients, the observed difference between NoAF and POAF values is very small (0.4 mmol/L) and is unlikely to be clinically meaningful. Moreover, both of these mean values are well within normal limits, and may not be pathophysiologically significant.

Other groups have done similar studies with other ions to varying degrees of success. For instance, in patients who have undergone cardiac surgery, hypokalemia has been associated with increased risk of POAF^10,11. In the Study of Prevention of Post-Operative Atrial Fibrillation (SPPAF) trial, the incidence of POAF in patients with serum potassium of 3.9 mmol/L or less was 50.7%, and 32.9% in those with potassium or 4.4 mmol/L or higher¹². Physiologically, hypokalemia reduces the resting potential, hastens depolarization, and increases automaticity and excitability¹³.

Deficiencies in magnesium are also known to be associated with arrhythmias. However, there have been mixed results on whether administration of magnesium in the post-operative period provides benefits. While some analyses have shown that prophylactic magnesium administration reduces the incidence of POAF, average length of hospital stay, and provides mortality benefits^14,15, others have either shown that administration does not alter length of hospital stay or mortality¹⁶, or reduce the incidence of POAF at all^17,18.

The role of calcium and chloride are less clear. While, blockade of the L-type calcium channels reduces intracellular calcium entry, formation of the calcium-troponin complex, and subsequently induces vasodilation and depresses myocardial contractility,¹⁹ no studies to date have examined the direct relationship between serum calcium levels and the risk of POAF. One study found that post-operative serum chloride concentration was higher in patients who went on to develop POAF, with a mean of 111.91 mEq/L vs. 105.17 mEq/L²⁰. Nonetheless, further studies are needed to elucidate their roles in modulating cardiac conduction in the post-op setting.

To our knowledge, our study is the first that considers the five electrophysiologically relevant ions together. This is important given recent basic science studies that suggest ions in combination may synergistically modulate cardiac conduction and arrhythmia risk as well as mechanical function^21-25. Our data demonstrate that neither the initial post-operative BMP nor the rate of change of electrolytes over the course of post-operative period are effective in predicting POAF. These results suggest that these electrolytes are not independent predictors of post-operative atrial fibrillation. Rather, there may be one or more clinical variables that are simultaneously associated with both electrolyte imbalance and the development of POAF.

One other interesting finding we observed is that post-operative Na⁺ and Cl⁻ change in parallel with one another. One potential explanation for this phenomenon is that serum Na⁺ and Cl⁻ in the post-operative setting is dictated by the quantity of normal saline administered, as no other ions followed a similar trend. This pattern is also difficult to explain physiologically, as the Na⁺-Cl⁻ co-transporter in the kidneys accounts for only 5-15% of all glomerular filtrate²⁶. From descriptive analyses of our population, we find that advanced age, valvular procedures, increased cross-clamp and cardiopulmonary bypass time, non-smoker status, a history of hypertension, and having contraindications to beta blockers may be associated with higher risk for POAF. Physiologically, advanced age, valvular procedures, and hypertension are logical. Advanced age and hypertension can result in cardiac fibrosis and remodeling, and valvular procedures are generally more complex procedures that result in greater trauma to the heart. In addition, patients who undergo valvular procedures such as those with mitral and tricuspid valve issues will frequently have dilated atria rendering them more prone to developing POAF. A number of prior studies have noted that increased cross-clamp and cardiopulmonary bypass time are associated with increased incidence of POAF^27-29. Physiologically, increased duration of cross-clamp time and cardiopulmonary bypass time results in increased ischemia, which leads to increased risk for reperfusion injury. This leads to increased inflammation that can precipitate atrial fibrillation. Beta blocker contraindications in this population group may include history of COPD, decompensated heart failure, symptomatic bradycardia, heart block, or medication interactions. These comorbidities can lead to an expected increase in risk for post-operative complications. However, the smoking status association seen in this study is unusual. Smoking is known to be associated with hypertension as a result of causing endothelial dysfunction and atherosclerosis, which from our study is shown to be associated with POAF. Moreover, nicotine has been shown to induce atrial fibrosis, which is a known predisposing factor for arrhythmia³⁰. The exact reason why smoking is associated with lower risk of POAF in this study is unclear. One possibility is that “never smokers” and “former smokers” in the data set are both combined into the “no group.” As the time that former smokers quit is not specified, it may be the case that some in the non-smoker group are actually high risk if they only recently quit.

The ultimate goal of this work is to identify a simplified POAF signature that would allow physicians to easily estimate a patient’s risk of developing POAF. At present, our data suggest that electrolyte panels alone cannot predict POAF. Interestingly, the random forest models effectively classified patients at low risk for POAF, likely because of the significant number of true negatives relative to the few incorrectly classified false positive cases. To successfully identify discrete high or low-risk patients, additional multivariate models will need to be constructed that simultaneously examine the role of electrolytes with other risk factors such as those presented in our table of descriptive statistics. Or, it may be worthwhile to restrict the study to patients at a high risk for POAF, such as those undergoing combination CABG and valve replacement surgery. Additional patient data may also be necessary to confirm the generalizability of the model to other populations beyond the patients included in this study. Ultimately, from these results, physicians may be able to react appropriately given a patient’s estimated risk, and to minimize adverse outcomes related to POAF.

Study Limitations

One potential limitation is that this study was solely performed in patients who had undergone cardiac surgery. These patients often have additional risk factors for the development of POAF as a result of both their procedure and medical history that other patients may not. As a result, the findings in this investigation may not be generalizable to other post-operative populations. Another limitation is that we used the STS database in correlation with medical records rather than using the medical record alone. This introduces limitations of using a database. This includes the potential that the database could be incomplete and thus missing some patients who truly had POAF or vice versa that patients could have been included who may have had benign rhythms but were inappropriately classified as POAF.

We chose to utilize logistic regression and random forest as our predictive models, both of which are optimized for our data set. However, it may be possible that conventional linear statistics may not reveal all electrolyte patterns. For instance, while our models were ultimately unable to predict patients at high risk for POAF, they notably demonstrated high specificities and negative predictive values, suggesting underlying populations at low risk for POAF. However, we were unable to determine discrete electrolyte ranges that corresponded with low risk for POAF. Future studies may be able to apply non-linear, multidimensional statistics to potentially elucidate new trends in the data.

One other limitation is that we did not account for the effect of external variables on electrolyte levels. For instance, many cardiac surgery patients take medications such as anti-arrhythmics, anti-hypertensives, insulin, and diuretics which may alter serum ionic balance. Other variables such as acute kidney injuries, hypoalbuminemia, and fluid shifts in the post-operative setting can also affect electrolytes. Future studies can attempt to account for their effect by including them as variables in the predictive models.

Finally, we did not conduct any analyses with the pre-operative BMP. As a result, we may not have accounted for any impact the procedure itself had on electrolytes, or any affect the rate of change of electrolytes may have had on the development of POAF. It may be worthwhile to conduct similar analyses with the pre-operative BMP in the future to identify patients at risk prior to proceeding with the operation.

Conclusion

Ionic imbalance has long been examined as a substrate for the development of post-operative atrial fibrillation. While absolute deviations in sodium and potassium are known predisposing factors for arrhythmia, the results of this study suggest that there appears to be questionable value in evaluating these ions in isolation to predict POAF. However, we do advocate for continued monitoring of electrolytes in the post-operative setting, as the addition of other risk factors in the context of ionic imbalance may be able to precipitate POAF.

Figure 1:

Flowchart of patients who met inclusion criteria