Introduction
Atrioventricular septal defects are a congenital malformation of the heart in which there is a common atrioventricular junction with a single atrioventricular valve. There are generally concomitant deficiencies in the atrioventricular septum leading to an atrial and/or ventricular level shunt. Reference Jacobs, Franklin and Béland1,Reference Jacobs, Franklin and Béland2 The surgical repair of an atrioventricular septal defect entails dividing the common atrioventricular valve into separate left and right atrioventricular valves and eliminating the atrial and/or ventricular level shunts. This can be done using a single-patch or two-patch technique. Reference Karl, Provenzano, Nunn and Anderson3–Reference Malik and Khalid Nuri6
As left atrioventricular valve regurgitation is generally amongst the frequent acute issues, postoperatively, there are institution-specific or surgeon-specific haemodynamic goals to preserve the integrity of the left atrioventricular valve after repair. Reference Ahmed, Alahmadi and Anjum7,Reference Farag, Isern Hacker and Grieshaber8 There is a paucity of data regarding this.
The primary aim of this study was to characterise the association of postoperative haemodynamics in the first 48 h with the progression of left atrioventricular valve regurgitation after complete repair of an atrioventricular septal defect in infants.
Methods
Study design
This study was a single-centre, retrospective study utilising high-fidelity physiologic data aggregated using the Sickbay platform (Medical Informatics Company, Houston, TX). The primary aim of this study was to characterise the association of postoperative haemodynamics in the first 48 h with progression of left atrioventricular valve regurgitation after complete repair of an atrioventricular septal defect in infants. Reference Jacobs, Franklin and Béland1,Reference Jacobs, Franklin and Béland2
This study received institutional review board approval and is in concordance with the Helsinki Declaration.
Patient selection
Paediatric patients under 1 year of age cared for in the paediatric cardiac ICU after an atrioventricular septal defect between January 1, 2022, and May 1, 2025, were eligible for inclusion. All varieties of atrioventricular septal defect were included. Those with double outlet right ventricle or other significant congenital malformations of the heart were excluded. Those who required mechanical circulatory support postoperatively, those without Sickbay data captured, and those without an arterial line were also excluded. Those without an arterial line were excluded from this study, as noninvasive blood pressure measurement is not done at the same fidelity as invasive arterial blood pressure measurement is, impacting the type of analyses utilised.
Variables of interest
The following patient-level data were captured: age at the time of surgery, weight at the time of surgery, type of repair, and discharge disposition. Data for the following haemodynamic variables were captured: heart rate, mean arterial blood pressure, central venous pressure, arterial saturation, respiratory rate, end-tidal carbon dioxide, cerebral near-infrared spectroscopy, and renal near-infrared spectroscopy.
Haemodynamic data were captured using the Sickbay platform, which captures these data every 1 s from the source devices. Data were captured for the first 48 h postoperatively. For this study, Sickbay data was captured as follows: heart rate in beats per minute from telemetry, mean arterial blood pressure in mmHg from an arterial line, central venous pressure in mmHg from a central line, arterial saturation in percent from a pulse oximeter, respiratory rate in breaths per minute from a pulse oximeter, end-tidal carbon dioxide in mmHg, cerebral near-infrared spectroscopy in percent from an INVOS 5100 system (Medtronic, Boston, MA), and renal near-infrared spectroscopy in percent from an INVOS 5100 system (Medtronic, Boston, MA).
Echocardiographic data was collected from the postoperative transesophageal echocardiogram and the discharge echocardiogram. Progression from trivial or mild left atrioventricular valve regurgitation to moderate or severe regurgitation between echocardiograms was considered worsening left atrioventricular valve regurgitation for the purposes of this study. These data were collected from the report itself. Institutionally, the degree of atrioventricular valve regurgitation is assessed using the width of the vena contracta. Descriptors such as mild to moderate or mild plus led to the classification being the more severe of the two such that “mild to moderate” was recorded as “moderate” and “mild plus” as “moderate.”
Operative technique
Our modified single-patch technique has been reported earlier. Reference Backer, Kaushal and Mavroudis9 After sternotomy and establishing cardiopulmonary bypass with bicaval venous cannulation, a vent is placed in the right superior pulmonary vein. After administration of cold blood cardioplegia to obtain cardiac standstill, a medial right atriotomy is performed. The common atrioventricular valve is floated by injecting saline into the ventricles, and the midportion of the common atrioventricular valve, where it will be separated into right and left components, is established. Next, a series of pledgeted sutures is placed on the right ventricular side of the ventricular septal crest. These sutures are passed through the dividing line between the common atrioventricular valve leaflets, which are not divided. The sutures are subsequently passed through an autologous pericardial patch. The sutures are then tied, obliterating the ventricular septal component of the defect. Next, the left atrioventricular valve zone of apposition between the bridging leaflets is approximated. The pericardial patch is then used to close the atrial component of the atrioventricular septal defect. The right-sided atrioventricular valve is then reconstructed. The right atrium is closed, and de-airing manoeuvres are performed. After which the patient is separated from cardiopulmonary bypass, and modified ultrafiltration is performed.
Our two-patch technique for atrial septal defect closure has also previously been described. Reference Mavroudis and Backer10 Cardiopulmonary bypass and exposure of the atrioventricular septal defect are established in a similar fashion as for the single-patch technique. The distance between the superior aspect of the atrioventricular valve and the crest of the ventricular septum is assessed to place the valve at the appropriate height above the septum and to prevent subaortic stenosis. The superior bridging leaflet is frequently divided at its midpoint to improve exposure. A crescent-shaped polytetrafluoroethylene patch is then fashioned. The patch is sutured to the right side of the ventricular septal crest. Sutures are then passed through the patch, through the midpoint of the bridging leaflets and then through a pericardial patch. The remainder of the repair, such as left atrioventricular valve zone of apposition closure, ostium primum atrial septal defect closure, and right atrioventricular valve reconstruction, is carried out as described for the single-match technique. The heart is de-aired, after which the patient is weaned from cardiopulmonary bypass.
The modified single-patch technique is utilised in patients in whom the ventricular component of the atrioventricular septal defect is not as large. The selection of patients for modified single-patch technique is described in greater detail in the previously referenced publications. If moderate or severe left atrioventricular valve regurgitation is present on the transesophageal echocardiogram at the end of the operation and there is a distinct modifiable area that can be addressed, then the patient is placed back on cardiopulmonary bypass for revision. If imaging does not demonstrate something modifiable, then the patient is not placed back on cardiopulmonary bypass for revision.
An atrial level shunt is not routinely left.
Statistical analyses, time-series plots
Time-series plots were generated to visualise the trend of each haemodynamic parameter over time. Each plot had two separate graphs, one for those who experience worsening left atrioventricular valve regurgitation and those who did not experience worsening left atrioventricular valve regurgitation. Plotted data were smoothened for non-overlapping 60 time periods.
Statistical analyses
Generalised estimating equation models with Gaussian distributions and identified link functions were used to evaluate associations between postoperative physiologic trajectories and worsening left atrioventricular valve regurgitation. This methodology was selected to account for repeated within-patient physiologic measurements while estimating population-averaged associations. In the context of a limited number of the events of interest, this methodology was felt to be able to provide robust estimation of regression coefficients while accounting for within-patient correlation without requiring estimation of patient-specific random effects, which may be unstable in small samples with limited outcome events. Robust sandwich variance estimates were used to obtain standard errors that were less sensitive to misspecification of the within-subject correlation structure.
Separate generalised estimating equation models were fit for each haemodynamic variable. For each model, postoperative time and worsening left atrioventricular valve regurgitation status were entered as independent variables. Postoperative time was modelled using restricted cubic spline basis functions with 3 degrees of freedom to allow for nonlinear physiologic changes during the postoperative period while avoiding overparameterisation. This spline specification was selected a priori to balance flexibility in trajectory estimation against overfitting risk in a small cohort. An exchangeable working correlation structure was specific to account for repeated measurements within patients. The null hypothesis for each model was that worsening left atrioventricular valve regurgitation was not associated with postoperative haemodynamic variables after accounting for postoperative time.
Given the modest sample size and limited number of events of interest, multivariate adjustment beyond postoperative time and worsening valve regurgitation was intentionally minimised to reduce model instability and overfitting.
It is important to note that these models ultimately answer if patients who ultimately have worsened left atrioventricular valve regurgitation have statistically higher or lower haemodynamic trajectories compared to those who do not.
All analyses were done on a local Jupyter Python environment. A p-value of less than 0.05 was considered statistically significant. Any use of the word “significant” or “significantly” in this manuscript will refer to statistical significance unless explicitly stated to be clinical significance.
Results
Cohort characteristics
A total of 28 patients were included in the final analyses. Haemodynamic data was collected for 4,838,400 timepoints. Nine (32.1%) of the patients had worsening left atrioventricular valve regurgitation. There was no significant difference in age, weight, or hospital length of stay between those with and without worsening left atrioventricular valve regurgitation. For the entire cohort, there was only 1 (3.6%) inpatient mortality. Only 3 (10.7%) of patients had moderate left atrioventricular valve regurgitation on the final postoperative transesophageal echocardiogram, while 12 (42.8%) had moderate or severe regurgitation on their discharge echocardiogram. Of the 28 patients, 13 (46.4%) had a one-patch repair while 15 (53.6%) had a two-patch repair.
Heart rate
Heart rate trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 1. Initial modelling without any other haemodynamic variables demonstrated that a higher heart rate was significantly associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of heart rate over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
Mean arterial blood pressure
Mean arterial blood pressure trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 2. Initial modelling without any other haemodynamic variables demonstrated that higher mean arterial blood pressure was significantly associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of men arterial blood pressure over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
Mean arterial blood pressure was also analysed in two other ways. The first was looking at the time integral of mean arterial blood pressure over time.
The second was looking at mean arterial blood pressure variability. This was done utilising the standard deviation of mean arterial blood pressure.
Central venous pressure
Central venous pressure trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 3. Initial modelling without any other haemodynamic variables demonstrated that higher central venous pressure was significantly associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of central venous pressure over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
Arterial saturation
Arterial saturation trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 4. Initial modelling without any other haemodynamic variables is demonstrated in Figure 4. Initial regression analyses without any other haemodynamic variables demonstrated that lower arterial saturations were associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of arterial saturation over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
Respiratory rate
Respiratory rate trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 5. Initial modelling without any other haemodynamic variables demonstrated that a higher respiratory rate was significantly associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of end-tidal carbon dioxide over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
End-tidal carbon dioxide
End-tidal carbon dioxide level trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 6. Initial modelling without any other haemodynamic variables demonstrated that a higher end-tidal carbon dioxide level was associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of cerebral near-infrared spectroscopy over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
Cerebral near-infrared spectroscopy
Cerebral near-infrared spectroscopy level trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 7. Initial modelling without any other haemodynamic variables demonstrated that a higher cerebral near-infrared spectroscopy level was associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Graph of renal near-infrared spectroscopy over time in those with and without worsening left atrioventricular valve regurgitation. The red line represents those without, and the blue line represents those with worsening left atrioventricular valve regurgitation.
Renal near-infrared spectroscopy
Renal near-infrared spectroscopy trends for those with and without worsening left atrioventricular valve regurgitation are demonstrated in Figure 7. Initial modelling without any other haemodynamic variables demonstrated that a lower renal near-infrared spectroscopy value was associated with worsening left atrioventricular valve regurgitation (p < 0.01).
Generalised estimation equation modelling
Generalised estimation equation modelling demonstrated no significant association of worsening left atrioventricular valve regurgitation with heart rate (beta coefficient 5.7, p = 0.21). There was no significant association with time.
There was a significant association of worsening left atrioventricular valve regurgitation with mean arterial blood pressure (beta coefficient 3.1, p = 0.04). There was no significant association with time.
There was a significant association of worsening left atrioventricular valve regurgitation variability with mean arterial blood pressure (beta coefficient 1.3, p = 0.03). There was no significant association with time.
There was no significant association of worsening left atrioventricular valve regurgitation with central venous pressure (beta coefficient 1.44, p = 0.34) or time.
There was no significant association of worsening left atrioventricular valve regurgitation with arterial saturation (beta coefficient 0.6, p = 0. 43). There was no significant association with time.
There was no significant association of worsening left atrioventricular valve regurgitation with respiratory rate (beta coefficient 1.8, p = 0.36). There was a significant association with time (p = 0.01).
There was no significant association of worsening left atrioventricular valve regurgitation with end-tidal carbon dioxide level (beta coefficient 4.3, p = 0.11).
There was a significant association of worsening left atrioventricular valve regurgitation with cerebral near-infrared spectroscopy (beta coefficient −2.1, p = 0.03). There was no significant association with time.
There was a significant association of worsening left atrioventricular valve regurgitation with renal near-infrared spectroscopy (beta coefficient −2.7, p = 0.04). There was no significant association with time.
Discussion
This study demonstrates that worsening left atrioventricular valve regurgitation (trivial or mild to moderate or severe) from immediately postoperatively to discharge after complete repair of an atrioventricular septal defect occurred in 32.1% of infants. Multivariate regression analysis demonstrated that increasing mean arterial pressure, increasing variability in mean arterial pressure, and decreasing renal near-infrared spectroscopy in the first 48 h postoperatively are associated with increased probability of worsening left atrioventricular valve regurgitation. With respect to mean arterial blood pressure, it appears to be a greater postoperative fluctuation rather than the overall time-dependent dose of mean arterial blood pressure that is associated with worsening left atrioventricular valve regurgitation.
These are novel findings, as no previous study to datehas been published associating postoperative haemodynamics with a worsening left atrioventricular valve. This comes in the setting of anecdotal concern from cardiothoracic surgeons regarding postoperative haemodynamics and left atrioventricular valve regurgitation. Somewhat arbitrary goals for haemodynamic variables are set postoperatively with hopes of minimising the progression of left atrioventricular valve regurgitation.
The findings that increased mean arterial blood pressure (with specific attention to blood pressure variability), decreased cerebral near-infrared spectroscopy, and decreased renal near-infrared spectroscopy are associated with worsening left atrioventricular valve regurgitation must be interpreted in nuanced fashion. These are simply associations, and causation cannot be inferred from these findings. While the statistical tools used in these analyses are able to help with causal inference, the temporal resolution of the echocardiographic data does not allow for causation to truly be inferred. A state of high mean arterial pressure and low near-infrared spectroscopy in the setting may be the result of left atrioventricular valve regurgitation, but the study uses echocardiographic assessment before and after the time period that the physiologic data are collected, which makes it more likely that these are likely the result. But this cannot be definitively stated, as we do not have echocardiographic data from multiple timepoints to determine precisely when changes in left atrioventricular valve regurgitation happened. This would require continuous, or at least extremely frequent, echocardiographic assessment of the left atrioventricular valve regurgitation.
It is also necessary to address that some of the directions of the associations changed from univariable to multivariable analyses. This can certainly happen and demonstrates the need for multivariable analyses, as adjusting for other covariables can demonstrate that the direction of an association can be significantly altered due to collinearity, confounding, or suppression effects. This, in fact, is the reason multivariable analyses are necessary. From a pragmatic standpoint, the univariable analyses should be used to guide actual bedside monitoring, but the multivariable analyses represent why those individual states when present simultaneously are leading to the physiology that they are.
As such, it is difficult to know whether a significant change in haemodynamic status contributed to the worsened left atrioventricular valve regurgitation or if worsened left atrioventricular valve regurgitation contributed to the haemodynamic change. Overall, the changes found associated with worsening left atrioventricular valve regurgitation seemed to represent a state of elevated and fluctuating mean arterial blood pressure. As mean arterial blood pressure is the product of systemic flow and systemic vascular resistance, the increased mean arterial blood pressure could be the result of either increasing or decreasing. As central venous pressure, cerebral near-infrared spectroscopy, and renal near-infrared spectroscopy decreased concomitantly with the increase in mean arterial blood pressure, it appears that the overall physiologic situation is consistent with decreased systemic flow. It may be hypothesised, then, that an increase in systemic vascular resistance may have contributed to worsening left atrioventricular valve regurgitation, which then resulted in decreased systemic flow. While the individual haemodynamic changes are objective observations, the mentioned hypothesis is subjective.
Variability in mean arterial blood was significantly associated with worsening left atrioventricular valve regurgitation when compared to absolute mean arterial pressure or the time integral of mean arterial blood pressure, suggesting that fluctuations in mean arterial blood pressure over the course of the first 48 h postoperatively were associated with increased probability of worsening left atrioventricular valve regurgitation. It is feasible that fluctuations in mean arterial blood pressure cause more mechanical stress on the valve than a constant, high pressure. Such a phenomenon has been noted with aortic dissection as well. Reference Howard, Sideso, Handa and Rothwell11–Reference Erbel, Aboyans and Boileau13 Blood pressure variability and not absolute blood pressure have also been associated with stroke, cardiovascular events, and renal injury. Reference Rothwell, Howard and Dolan14 From a mechanical perspective, variable blood pressure exposes the left atrioventricular valve and its support structures to repeated, unsteady pressure gradients and dynamic loads. These fluctuations may overwhelm the compensatory mechanisms of a repaired valve, leading to transient or progressive loss of coaptation and worsening regurgitation. This is supported by principles of fluid dynamics and mechanical fatigue, where structures subjected to cyclical, variable stress are more prone to dysfunction and failure than those under constant load. Reference Sacks and Yoganathan15–Reference Yoganathan, He and Casey Jones17 These findings reflect dynamic modulation of regurgitation severity rather than structural valve deterioration over a short postoperative interval.
The results demonstrate that lower mean arterial blood pressure, less fluctuating mean arterial blood pressure, higher cerebral near-infrared spectroscopy, and higher renal near-infrared spectroscopy represent haemodynamic associations with which the probability of worsening left atrioventricular valve is minimised. As has been demonstrated in other circulatory physiologies, a high cardiac output, low systemic vascular resistance state appears optimal to minimise worsening of left atrioventricular valve regurgitation after complete repair of a complete atrioventricular septal defect repair in infants.
This study leverages high-fidelity physiologic data, with over 4 million timepoints captured across patients. The high temporal resolution allows for precision characterisation of physiologic trajectories and reduces measurement noise in time-varying covariates within each patient. However, the primary outcome was defined at the patient level, and therefore, the effective sample size for outcome modelling is determined by the number of patients rather than the number of timepoints. As such, while the high-fidelity data strengthen the estimation of physiologic patterns, the relatively small number of patients limits statistical power for between-patient comparisons.
Importantly, from a clinical standpoint, worsening left atrioventricular valve regurgitation in the postoperative period is not determined by physiologic conditions at a single discrete timepoint but rather by the cumulative and dynamic physiologic environment over time. Clinicians do not make decisions based solely on an isolated measurement at hour 2 or hour 24 but instead interpret evolving trends in heart rate, blood pressure, central venous pressure, and markers of systemic oxygen delivery over the postoperative course. Accordingly, the analytic approach in this study was designed to reflect this clinical reality by incorporating measured physiologic data across the 48-h postoperative period rather than restricting it to discrete timepoints. This allows for characterisation of physiologic trajectories and variability, which may be more relevant to valve function rather than isolated measurements. As such, the associations identified in this study should be interpreted as reflecting physiologic patterns over time rather than time-specific effects. The two echocardiograms utilised for quantifying left atrioventricular valve regurgitation flank the study period, and thus, one is before, and one is after.
Despite the strengths of this study, this study is not without its limitations. This is a single-centre retrospective study, and centre-specific variability may limit the generalisability of its findings. For instance, variations in operative technique could not be largely captured across a relatively small number of surgeons, although the type of repair (one or two patches) was accounted for. Additionally, while the analyses are statistically well powered due to the large number of datapoints, the generalisability is limited by the number of patients. Additionally, there is an even smaller number of patients who meet the endpoint of interest, which, in this case, is worsening left atrioventricular valve regurgitation. Thus, these data do not imply causation or prediction by association. Additionally, specific data regarding vasoactive inotrope score, end organ function, fluid balance, mechanical ventilation, and sedation were not included. This was done intentionally to look at the association with the raw vital sign data. These data should be viewed as hypothesis-generating. Although commonly cited heuristics such as 10 events per variable are often used in regression modelling, such thresholds are not universally applicable, and model adequacy depends on multiple factors, including underlying signal strength and predictor structure. Reference van Smeden, de Groot and Moons18,Reference Vittinghoff and McCulloch19 Nevertheless, given the relatively small number of patients and outcome events in this study, findings should be interpreted as hypothesis-generating. Additionally, these findings are limited to only the operative admission and should not be applied to longer periods of time. These are preliminary analyses, and future analyses with larger sample sizes and more data are critical.
Conclusion
Mean arterial blood pressure, cerebral near-infrared spectroscopy, and renal near-infrared spectroscopy values in the 48 h following a complete repair of an atrioventricular septal defect in infants are independently associated with worsening left atrioventricular valve regurgitation. With respect to arterial blood pressure, both absolute values and variability are associated with worsening atrioventricular valve regurgitation.
Open access
