Screening for cardiac allograft rejection among heart transplant recipients using an electrocardiogram-based deep learning model

Background

Current approaches utilizing non-invasive methods to screen for cardiac allograft rejection (gene expression profiling and cell free DNA) have yet to be broadly integrated into heart transplant management and have shown limited discrimination (AUCs of 0.68 and 0.77, respectively). Changes in the electrocardiogram (ECG) have been reported at the time of severe cardiac rejection, including low voltages and conduction abnormalities. It remains unknown if subtle ECG changes correlating with cardiac allograft rejection can be detected earlier using machine learning methods.

Purpose

We sought to develop an artificial intelligence (AI) model to detect cardiac allograft rejection based on the 12 lead ECG.

Methods

We identified all patients who underwent a heart transplant at 3 hospital sites within a single health system from Jan 1998 through Apr 2021 and extracted digital 12-lead ECG data as well as endomyocardial biopsy pathology results from the electronic medical record. We partitioned our data into a training (80%), validation (10%), and test set (10%) based on a group stratification sampling. Each patient was present in only one set and each set had a positivity rate of 2.6% with 6,074/758/758 ECGs belonging to 1,146/140/141 unique patients in each set respectively. Cardiac allograft rejection was defined as moderate or severe acute cellular rejection based on International Society for Heart and Lung Transplantation (ISHLT) guidelines. A convolutional neural network, using the 12-lead ECG data as input, was trained with hyperparameter optimization for regularization, learning rate adjustments, and class weights. Model performance metrics were based on the test data and estimated using the final model architecture.

Results

1,587 heart transplant recipients who had at least one endomyocardial biopsy were evaluated for inclusion. We limited our sample to ECGs performed within 30 days of the biopsy date (7,590 ECGs, representing 1,425 unique patients). Our study population had a median age of 55.8 years and 28.7% were female. The median number of ECG-biopsy pairs per patient was 5. The majority of endomyocardial biopsy results were classified as none or mild rejection (97.1%), and 2.9% had moderate/severe rejection. The ECG-based AI model detected cardiac allograft rejection with an area under the receiver operative curve (AUC) of 0.84 in the test set. The sensitivity, specificity, positive and negative predictive values were 95%, 52.6%. 5.2% and 99.7% respectively.

Conclusions

An AI-ECG model appears to outperform novel non-invasive laboratory tests (gene expression profiling and cell free DNA) for detecting cardiac allograft rejection and does not require a blood draw or the additional complexities surrounding sample processing. This model relies on a readily available and relatively inexpensive test, the ECG. In addition, AI predictions can be made available within a few minutes following ECG acquisition.

Funding Acknowledgement

Type of funding sources: Private hospital(s). Main funding source(s): Mayo Clinic

Donor cerebral tissue pulmonary emboli in a functioning transplanted lung

Cerebral tissue pulmonary embolus (CTPE) is a rare event following severe blunt or penetrating head trauma and is often complicated by coagulation disturbances and hemorrhage. Donor cerebral tissue pulmonary embolism has been reported to cause lethal, early graft dysfunction in lung transplant recipients. We report a case of donor cerebral tissue pulmonary embolism in a 41-year-old female single lung transplant recipient with excellent post-operative graft function.

Differences in rejection grading after simultaneous pancreas and kidney transplantation in pigs

Clinical observations suggest that recipients of multiorgan transplants from the same donor can have disparate immunological reactions to each organ. We studied this phenomenon in 36 diabetic (streptozotocin-induced), bilaterally nephrectomized, immunosuppressed (cyclosporine, azathioprine, prednisone) pig recipients of simultaneous (same donor) pancreas (bladder drained) and kidney allografts by grading the histological intensity of rejection in biopsies of each organ at defined intervals posttransplant. Graft function was monitored by plasma glucose (PG) and urine amylase (UA) for the pancreas and serum creatinine (Cr) for the kidney. Interstitial rejection was graded as absent, mild, moderate, and severe in, respectively, 8%, 25%, 42%, and 25% of pancreas vs. 4%, 12%, 27%, and 57% of kidney biopsies at 1 week; and 0%, 43%, 29%, and 29% of pancreases vs. 10%, 0%, 30%, and 60% of kidneys at two weeks. Although the distribution of grades was similar in the two organs (P > 0.1), the grade of rejection for each pair at 1 week (n = 24) was discordant in 75% (42% differed by one and 33% by > or = 2 grades) and at 2 weeks (n = 7) in 57% (29% by 1 and 29% by > or = 2 grades). The inability to use the severity of interstitial rejection in one organ to predict the findings in the other is exemplified by the fact that for the two pancreases without interstitial rejection at one week, the corresponding kidney showed moderate or severe rejection, and for the 1 kidney without rejection the corresponding pancreas showed moderate rejection. Vascular rejection grades (absent, mild, moderate, severe) also showed a similar distribution for the pancreas (57%, 30%, 9%, 4%) vs. kidney (50%, 38%, 0%, 12%) at 1 week, and at 2 weeks (57%, 29%, 0%, and 14% for the pancreas vs. 78%, 11%, 0%, and 11 for the kidney) (P > or = 0.64). However, the grading of vascular rejection in organ pairs was dyssynchronous in 54% at 1 week (n = 22) and 29% at 2 weeks (n = 7). No vascular rejection in the pancreas with rejection in the kidney was seen in 5 pairs at 1 week (23%) and 0 at 2 weeks (0%), while no rejection in the kidney with rejection in the pancreas was seen in 5 pairs at 1 week (23%) and 2 pairs at 2 weeks (29%).(ABSTRACT TRUNCATED AT 400 WORDS)

Differences in rejection grading after simultaneous pancreas and kidney transplantation in pigs

Clinical observations suggest that recipients of multiorgan transplants from the same donor can have disparate immunological reactions to each organ. We studied this phenomenon in 36 diabetic (streptozotocin-induced), bilaterally nephrectomized immunosuppressed (cyclosporine, azathioprine, prednisone) pig recipients of simultaneous (same donor) pancreas (bladder drained) and kidney allografts by grading the histological intensity of rejection in biopsies of each organ at defined intervals posttransplant. Graft function was monitored by plasma glucose (PG) and urine amylase (UA) for the pancreas and serum creatinine (Cr) for the kidney. Interstitial rejection was graded as absent, mild, moderate, and severe in, respectively, 8%, 25%, 42%, and 5% of pancreas vs. 4%, 12%, 27%, and 50% of kidney biopsies at 1 week; and 0%, 43%, 29%, and 29% of pancreases vs. 10%, 0%, 30%, and 60% of kidneys at two weeks. Although the distribution of grades was similar in the two organs (P > 0.1), the grade of rejection for each pair at 1 week (n = 24) was discordant in 75% (41% differed by one and 35% by > or = 2 grades) and at 2 weeks (n = 7) in 57% (29% by 1 and 29% by > or = 2 grades). The inability to use the severity of interstitial rejection in one organ to predict the findings in the other is exemplified by the fact that for the two pancreases without interstitial rejection at one week, the corresponding kidney showed moderate or severe rejection, and for the 1 kidney without rejection the corresponding pancreas showed moderate rejection. Vascular rejection grades (absent, mild, moderate, severe) also showed a similar distribution for the pancreas (57%, 30%, 9%, 4%) vs. kidney (56%, 30%, 0%, 13%) at 1 week, and at 2 weeks (57%, 29%, 0%, and 14% for the pancreas vs. 78%, 11%, 0%, and 11% for the kidney) (P > or = 0.64). However, the grading of vascular rejection in organ pairs was dyssynchronous in 51% at 1 week (n = 22) and 29% at 2 weeks (n = 7). No vascular rejection in the pancreas with rejection in the kidney was seen in 5 pairs at 1 week (23%) and 0 at 2 weeks (0%), while no rejection in the kidney with rejection in the pancreas was seen in 5 pairs at 1 week (23%) and 2 pairs at 2 weeks (29%).(ABSTRACT TRUNCATED AT 400 WORDS)

Rejection in single versus combined pancreas and kidney transplantation in pigs

Clinically, the incidence of reversible renal allograft rejection episodes appears to be higher in recipients of simultaneous pancreas/kidney (SPK) than of kidney transplantation alone (KTA); conversely, the rate of irreversible pancreas allograft rejection appears to be higher in pancreas transplant alone (PTA) than SPK recipients. Clinical/histological correlation of graft rejection in these three groups has not been precise. Therefore, we studied the incidence and histological severity of the rejection process in a large animal (pig) model of SPK (n = 36), PTA (n = 31), and KTA (n = 36) allotransplantation. SPK and PTA recipients were made diabetic pretransplant by streptozotocin (150 mg/kg). Pancreas graft exocrine secretions were bladder-drained via a duodenocystostomy for urine amylase (UA) monitoring; endocrine function was monitored by plasma glucose (PG) levels. SPK and KTA recipients underwent native nephrectomy, and renal allograft function was monitored by serum creatinine (CR). Cyclosporine, azathioprine, and prednisone were given in tapering doses from the time of transplantation. Grafts were biopsied weekly to grade histologic severity of interstitial and vascular rejection on light microscopy (LM) and for intensity of T cell infiltration on immunofluorescence. Pancreas graft exocrine function (UA above pretransplant baseline), present in 62% of PTA and 68% of SPK recipients at one week, persisted in only 7% of PTA vs. 64% of SPK pigs at 2 weeks (P = 0.0004). Likewise, pancreas graft endocrine function (PG < 200 mg/dl off insulin) was sustained longer in SPK than PTA recipients (100% vs. 84% at 1 and 91% vs. 27% at 2 weeks; P = 0.0006). However, renal allograft functional survival (serum creatinine < 3.0 mg/dl) was not significantly different (P = 0.471) between SPK and KTA recipients (36% vs. 30% at 1 and 23% vs. 13% at 2 weeks). Graft functional parameters partially correlated with biopsy observations. Pancreas allograft biopsies showed a significantly (P = 0.03 at 1 and P = 0.05 at 2 weeks) lower incidence of moderate/severe interstitial rejection in SPK than PTA recipients (67% vs. 95% at 1 and 57% vs. 92% at 2 weeks); rejection was absent in 8% of SPK and in no PTA biopsies at 1 week. Vascular rejection was moderate/severe in significantly fewer (P = 0.0013 at 1 and P = 0.023 at 2 weeks) SPK than PTA pancreas grafts (13% vs. 37% at 1 and 14% vs. 38% at 2 weeks).(ABSTRACT TRUNCATED AT 400 WORDS)

[Metabolic and immunologic sequelae of pancreas transplantation in the pig]

After induction of a Streptozotocin-induced diabetes mellitus solitary pancreas transplants were performed in 31 Yorkshire-Landrace pigs. Exocrine pancreatic secretions were bladder drained (duodenocystostomy). Postoperatively all animals were subject to a standardized immunosuppression. The purpose of the study was both to examine the intensity of rejection episodes within a certain time period and to correlate laboratory with histology data. Cystoscopic transduodenal pancreas biopsies in defined intervals were obtained by intraoperative ultrasound control; this technique reduces the complication rate and should be used clinically. The biopsy results showed that severe rejection episodes despite high dose triple-immunosuppression for induction therapy were noted histologically and immunohistochemically as early as postoperative day 6. Morphological changes significantly preceded functional changes. Due to the distinct immunogenicity of pancreas transplants and in order to decrease incidence and severity of rejection episodes, quadruple-immunosuppression (including T-cell-antibodies) is imperative for induction therapy and anti-rejection treatment. Our results demonstrate that solitary pancreas transplantation is clinically indicated only in patients with extremely labile diabetes mellitus due to the high immunosuppression required. This animal model should be utilized for further studies to improve the result of solitary pancreas transplantation.

Streptozotocin-induced diabetes mellitus in pigs

We induced, as a precondition for a pancreas transplant, insulin-dependent diabetes mellitus in 67 Yorkshire Landrace pigs by administering streptozotocin. A dosage of 150 mg/kg body weight gave rise to a long-lasting diabetes mellitus that persisted with time (follow-up period: 26 weeks). Consecutive measurements of serum glucose and plasma insulin, before and up to 30 hours after administering streptozotocin, revealed triphasic behavior: initial hyperglycemia (1st to 3rd hour), pronounced hypoglycemia (12th to 18th hour), then hyperglycemia (22nd hour on). IVGTTs done 1 to 7 days after administering streptozotocin revealed a reduction of the K-value (glucose disappearance rate) from 0.3 (day 2) to 0.07 (day 4). Immunohistochemical studies revealed a complete loss of all beta-cells, concomitantly with a relative increase in glucagon- and somatostatin-positive cells. We also observed a complete loss of pp (pancreatic polypeptide)-positive cells. Diabetes induced by streptozotocin at 150 mg/kg body weight is complete and permanent; our mortality rate was 0%. Given the high morbidity rate after pancreatectomy, streptozotocin should be the method of choice for inducing diabetes mellitus in pigs.

Mesenchymal cells isolated after acute lung injury manifest an enhanced proliferative phenotype

After acute lung injury, mesenchymal cells migrate into the alveolar airspace where they proliferate and deposit connective tissue macromolecules. Early in the disease process, inflammatory cell-derived trophic factors modulate these mesenchymal cell functions. However, in those patients who die, even as the inflammatory response abates, the fibroproliferative response continues, resulting in extensive intraalveolar fibrosis. We therefore hypothesized that lung mesenchymal cells obtained from individuals dying with acute alveolar fibrosis would manifest an enhanced proliferative capacity that was independent of persistent exogenous signals. To examine this hypothesis, the in vitro growth properties of mesenchymal cells prepared from patients dying with acute lung injury (n = 3) were analyzed in defined medium and compared with those of mesenchymal cells similarly prepared from patients dying with histologically normal lungs (n = 3). Isolates were characterized as mesenchymal cells by using morphological and immunohistochemical criteria. In accord with the hypothesis, mesenchymal cells isolated from lung-injured patients doubled within 3 d in the complete absence of exogenous peptide growth factors, reaching a saturation density of approximately 15 x 10(3) cells/cm2. As expected, lung mesenchymal cells from normal individuals failed to significantly increase in number. Consistent with this proliferative phenotype, the immediate early cell division cycle genes c-fos and c-jun were constitutively expressed in each cell strain prepared from injured lungs, but not in those from control lungs. The observed proliferative phenotype was stable through the fifth subcultivation of the cells. Despite these proliferative properties, three separate criteria indicated the mesenchymal cells from injured lungs were not transformed: normal karyotype; finite lifespan in vitro (9-10 subcultivations); and inability to disseminate in mice with severe combined immunodeficiency. These data support the hypothesis that mesenchymal cells manifest an enhanced proliferative state after acute lung injury.

Liver transplant rejection and cholestasis: comparison of technetium 99m-diisopropyl iminodiacetic acid hepatobiliary imaging with liver biopsy

To determine whether the scintigraphic evaluation of technetium-99m diisopropyl iminodiacetic acid (DISIDA) uptake and excretion can distinguish among liver transplant patients with biopsy evidence for rejection, cholestasis or neither condition, we reviewed scintigrams and biopsies in 36 patients. There were 76 scintigrams with corresponding biopsies. Uptake and excretion were graded from image data on scales reflecting normal through severely abnormal values. Biopsies were evaluated for findings of cholestasis and rejection. The majority of scintigrams demonstrated normal uptake (60/75, 80%) and delayed excretion (65/76, 85%), which was most marked immediately after transplantation. One-way analysis of variance showed that the mean excretion values significantly differed between patients with normal biopsies and those with cholestasis and/or rejection (P = 0.0003). However, mean uptake scores demonstrated no statistically significant difference between these two groups of patients (P = 0.1). These findings suggest that 99mTc-DISIDA scintigraphy can differentiate between transplants with and without rejection/cholestasis but not between rejection and cholestasis. If 99mTc-DISIDA excretion is normal, rejection and cholestasis are unlikely.

Use of epicardial electrocardiograms for detecting cardiac allograft rejection

Since the advent of cyclosporin A surface electrocardiograms have been unreliable for diagnosing cardiac allograft rejection. Although several noninvasive methods have been proposed, none have been sufficiently accurate to be considered for clinical use. We have studied the use of the QRS complex amplitude, the unipolar peak-to-peak amplitude, recorded from intramyocardial electrodes for detecting rejection. Ten adult mongrel dogs underwent placement of intramyocardial electrodes on each ventricle. After stabilization of signals the hearts were transplanted heterotopically into unmatched recipients receiving cyclosporin A, azathioprine and methylprednisolone. Endomyocardial biopsies were performed after stabilization of unipolar peak-to-peak amplitude, twice weekly thereafter, and when unipolar peak-to-peak amplitude fell significantly. This detected 13 of 14 episodes of rejection. There was one false-positive and one false-negative result. The false-negative study became positive the following day. Thus, analysis of unipolar peak-to-peak amplitude detected all episodes of rejection in a clinically relevant time frame and was able to detect mild forms of rejection and multiple episodes of rejection in the same heart even in the presence of therapeutic levels of cyclosporin A.

Allograft vasculopathy and death in a cardiac transplant patient with angiographically normal coronary arteries

A cardiac transplant patient died of severe left ventricular dysfunction 2 weeks after coronary angiography demonstrated normal coronary arteries. Post mortem examination showed significant and diffuse allograft vasculopathy. Coronary angiography may have limited sensitivity in detecting allograft vasculopathy.

Hereditary tyrosinemia type I (chronic form): pathologic findings in the liver

Hereditary tyrosinemia type I presents with either acute hepatic failure in the neonatal period or later in infancy with progressive liver dysfunction secondary to cirrhosis. The inevitably fatal outcome in those children with the chronic form has been transformed with the advent of liver transplantation. Native livers from five children who received allografts were studied pathologically and compared with earlier hepatic biopsies in two of these patients that had been performed several years before transplantation. Our findings support the conclusion that a sequence of morphologic changes from the initial micronodular cirrhosis through an intermediate mixed cirrhotic pattern to macronodular cirrhosis occurs. The micronodular phase is transitory, over a period of only a few months, since mixed micronodular macronodular cirrhosis was already present in the livers of children who received transplants by 11 months of age. Focal hepatocellular dysplasia was present in one of the livers with mixed cirrhosis but was not identified in the other two cases. Macronodular cirrhosis accompanied two cases of hepatocellular carcinoma in this study. In order to preclude the latter complication, liver replacement is necessary before the age of 2 years.

Heparin cofactor II assay. Elimination of heparin and antithrombin-III effects

Functional assays for heparin cofactor II (HC-II) are based on the inactivation of thrombin by HC-II in the presence of dermatan sulfate (DS). Residual thrombin is measured in a chromogenic assay. Interference by the antithrombin-III (AT-III)/heparin complex, which also rapidly inactivates thrombin, must be eliminated from the HC-II test system. Commercial DS is contaminated with heparin, while plasma specimens to be tested contain AT-III. After NaNO2/acetic acid treatment of DS (to inactivate heparin), there was enough residual heparin to cause AT-III interference. Treatment of plasma with commercially available anti-AT-III antiserum largely, but not completely, removed AT-III interference from the HC-II assay. With commercially available reagents, both NaNO2/acetic acid treatment of DS and anti-AT-III treatment of plasma were needed to eliminate heparin/AT-III interference. Protamine sulfate inactivated DS as well as heparin and could not be used to reduce AT-III/heparin interference with the HC-II assay.