Operative correction of tetralogy of Fallot frequently results in pulmonary insufficiency and chronic volume overload, which have been linked to increased risk for adverse outcomes. No consensus recommendations for the timing of pulmonary valve replacement (PVR) exist. The aim of this study was to examine the pattern of PVR in the United States from 2004 to 2012. The Pediatric Health Information Systems database was used to perform an observational study of children and adults ≥10 years of age with diagnoses of tetralogy of Fallot who underwent PVR at 35 centers in the United States from 2004 and 2012, to assess the rate of PVR and the age at which is performed. Mixed-effects multivariate regression was used to account for patient-level covariates and center-level covariance. Additional analyses assessed for trends in cost, hospital length of stay (LOS), intensive care unit LOS, and in-hospital mortality over the study period. In total, 799 subjects at 35 centers underwent PVR over the study period. The number of PVRs performed per year increased significantly over the study period. There was significant between-center heterogeneity in age at PVR (p <0.001). Age at PVR, intensive care unit LOS, hospital LOS, and cost did not change over the study period. In conclusion, PVR in patients with tetralogy of Fallot is being performed more frequently, without an accompanying change in the age at PVR or other measurable outcomes. There is significant variability in the age at which PVR is performed among centers across the United States. This highlights the need for additional research guiding the optimal timing of PVR.
The purpose of this study was to analyze trends in the rate of pulmonary valve replacement (PVR) in subjects with tetralogy of Fallot (TOF) at centers in the United States and the average age at PVR. Given the number of publications addressing the timing of PVR in patients with TOF during the study period, we hypothesized that the rate of PVR in subjects with TOF would increase with time and that this would be accompanied by decreasing age at PVR. We also sought to determine if changes in practice were accompanied by trends in perioperative outcomes.
Methods
The Pediatric Health Information Systems (PHIS) database is an administrative database that contains data from inpatient, emergency department, ambulatory surgery, and observation encounters from 43 not-for-profit, tertiary care pediatric hospitals in the United States, affiliated with the Children’s Hospital Association (Overland Park, Kansas). Data quality and reliability are assured through a joint effort between the Children’s Hospital Association and participating hospitals. The PHIS data warehouse is managed by Truven Health Analytics (Ann Arbor, Michigan). Participating hospitals provide discharge and encounter data, including demographics, diagnoses, and procedures. Forty-two of these hospitals submit resource utilization data (e.g., pharmacy products, radiologic studies, and laboratory studies) to PHIS. Data are deidentified at the time of submission and are subjected to reliability and validity checks before inclusion. A data-use agreement was signed between the study investigators and the Children’s Hospital Association. The institutional review board of The Children’s Hospital of Philadelphia reviewed the project and determined that it did not represent human subjects research, in accordance with the Common Rule (45 CFR 46.102[f]).
We included children and adults ≥10 years of age with diagnoses of TOF (International Classification of Diseases, Ninth Revision [ICD-9], code 745.2) who underwent either operative (ICD-9 codes 35.25 and 35.26) or transcatheter (ICD-9 code 35.07) PVR at any of the 43 PHIS centers from January 1, 2004, to December 31, 2012. The lower limit for age was chosen to reduce error in cohort identification, specifically the inclusion of subjects receiving other operations incorrectly coded as PVR, such as placement or replacement of right ventricle–to–pulmonary artery conduits. Centers were excluded if they performed <50 cardiac surgical procedures per year or <5 PVR procedures (all ages) over the study period or if they did not report cardiac surgical procedures in ≥66% (6 of 9 years) of years during the study period. This was intended to restrict analysis to centers with stable reporting practices and operative volumes.
Data were extracted from the PHIS database by direct query and included subject age, gender, race, insurance payer (private, public insurance, other), and presence of genetic syndrome (ICD-9) codes: 758.1, 758.2, 758.0, 758.31, 758.32, 758.33, 758.9, 758.6, 758.83, and 758.89. In-hospital death, length of stay (LOS), intensive care unit (ICU) LOS, and adjusted cost were also extracted. Center total cardiac surgical volume was calculated by admissions with ICD-9 codes 35.00 through 35.99 and 37.5, 37.51, 37.52, and 39.0 and 39.21.
Hospital charges are sent directly from billing records of each institution to PHIS. Cost is calculated by multiplying charge data by center-specific department-level cost-to-charge ratios, allowing comparison across centers at which charging practices vary. In this study, costs were standardized to 2012 United States dollars using the Consumer Price Index for medical care, compiled by the Bureau of Labor Statistics ( http://data.bls.gov/cgi-bin/dsrv ).
Descriptive statistics are expressed as mean ± SD, median (range and interquartile range), and percentages and counts as appropriate. We could not measure the total number of subjects with TOF during the study period and therefore could not calculate the incidence of PVR. The rate of PVR was assessed in 3 ways: (1) the total number of PVRs performed across included centers per year, (2) the average number of PVR operations per center per year (analyzed both with and without adjustment for center volume), and (3) the proportion of each center’s total cardiac surgical volume accounted for by PVR. Linear regression was used to assess for trends in each of these rates per year over the study period. The latter 2 measures of PVR rate were chosen to differentiate between changes in the rate of PVR and changes in cardiac surgical volume at centers over time. Although not the true incidence of PVR, the rates generated provide information about the trend in the number of PVR being performed. The association between date of PVR and age at PVR in years was assessed using mixed-effects models, on the basis of generalized linear models using normal or Gaussian distribution and canonical identity link. Fixed effects were included for prespecified covariates (listed previously). All covariates were included, without use of bivariate screening. To account for covariance within centers, a random intercept was added, and to account for covariance in the association between age of repair (outcome) and year of repair (exposure) by center, a random slope was added. Post hoc, additional models were tested to assess for nonlinearity in the association between year of operation and age at operation, including tests for changes in slope: (1) at the midpoint of the study period and (2) at the introduction of transcatheter PVR in the database (2011), as well as the addition of a quadratic term to assess for a nonlinear change in age with time. Goodness of fit was compared using likelihood ratio tests.
Association between year of PVR and secondary outcomes was assessed using generalized linear models with fixed effects and random effects. Risk for in-hospital death was modeled using binomial distribution. LOS, ICU duration, and cost-to-charge ratio–adjusted cost were modeled using log gamma distribution. For each model, the canonical link for the listed distribution was used. In addition to previously included covariates, age at PVR (centered at 16 years) was added to models as a covariate. Because the number of in-hospital deaths was small, multivariate models could not be used to assess risk factors for in-hospital death. For other outcomes, standardized values for outcomes of interest were determined using conditional standardization, holding fixed effects covariates equal to the referent group. Heterogeneity in age of repair by center was assessed using likelihood ratio test of random intercept term. Heterogeneity in the response in age of PVR over time by center was assessed using the likelihood ratio test for random slope term.
Sensitivity analyses were performed for the effect of various age inclusion and exclusion criteria on the results of the study. First, the lower limit on age was varied over a range of ages from 6 to 18 years, with primary analysis performed for each. This was performed to determine whether different age criteria were biasing study results with regard to temporal changes in age at PVR. A second sensitivity analysis for the upper limit of age was also performed restricting subject age from <60 to <25 years in 5-year increments. This was performed to ensure that high-age outliers were not exerting excessive influence on study results. Third, subjects who underwent transcatheter PVR were excluded to ensure that this subgroup of subjects was not exerting excessive influence on study results.
Introduction of transcatheter valve replacement was identified as a factor that might potentially have influence on study results. Several steps were taken to characterize these effects and account for them. First, the characteristics of subjects who underwent transcatheter and operative PVR were summarized. Comparisons between the transcatheter PVR (TC-PVR) and operative PVR (S-PVR) groups were made using either 2-tailed Fisher’s exact or Wilcoxon’s rank-sum tests as appropriate. Second, TC-PVR (compared with S-PVR) was included as a covariate in each multivariable model to condition for its effect. Third, sensitivity analyses of the primary model were performed to determine the influence of TC-PVR’s introduction on the age at PVR, specifically (1) including the year of the introduction of TC-PVR in the PHIS database (2011) as a potential hinge point in age of PVR and (2) restricting analyses to SC-PVR subjects.
All analysis was performed using Stata MP version 13 (StataCorp LP, College Station, Texas).
Results
From 2004 to 2012, a total of 799 subjects with TOF underwent PVR at 35 centers across the United States contributing data to the PHIS database ( Table 1 ). Eight centers (10 subjects) did not meet the inclusion criteria and were excluded. Median age at PVR was 17 years (range 10 to 64). Most of the population was male (57%) and white (75%). A small minority (10%) carried diagnoses of genetic syndromes. Median hospital LOS was 4 days (range 1 to 126), with a median duration of ICU stay 2 days (range 1 to 41), median cost of hospitalization of $44,029 (range $4,399 to $866,720), and in-hospital mortality of 0.9% (7 subjects).
| Number | 799 at 35 centers |
| Age at pulmonary valve replacement (years) | 17 (range: 10-64 IQR: 13-22) |
| Female Gender | 347 (43%) |
| Race | |
| White | 601 (75%) |
| Black | 78 (10%) |
| Asian | 27 (3%) |
| Other | 93 (12%) |
| Known genetic syndrome | 79 (10%) |
| Transcatheter pulmonary valve | 30 (4%) |
| Payor | |
| Private insurance | 467 (58%) |
| Medicaid | 213 (27%) |
| Other | 199 (16%) |
| Mortality | 7 (0.9%) |
| Cost (2012US$) ∗ | 44,029 (range: 4,399 – 866,720, IQR: 35,294 – 57,514) |
| Hospital length of stay (days) | 4 (range: 1-126, IQR: 3-5) |
| ICU stay (days) † | 2 (range: 1-41, IQR: 2-4) |
Characteristics of subjects who underwent SC-PVR and TC-PVR are listed in Table 2 . Subjects who underwent TC-PVR were older (p = 0.046) and more likely to be black (p = 0.04). Mortality was not significantly different (p = 1.00). Duration of ICU stay was not significantly different between TC-PVR and S-PVR (p = 0.21) but was available in a smaller percentage of subjects after TC-PVR than S-PVR (27% vs 84%, p <0.001). Cost was not significantly different, but hospital LOS was less after TC-PVR than S-PVR (median 1 day, range 1 to 14, interquartile range 1 to 2, p <0.0001).
| Operative valve replacement | Trans-catheter valve replacement | p | |
|---|---|---|---|
| Number | 769 at 35 centers | 30 at 18 centers | |
| Age at pulmonary valve replacement (years) | 17 (range: 10-61 IQR: 13-22) | 20.5 (range: 10-64, IQR: 15-28) | 0.046 |
| Female Gender | 333 (43%) | 15 (50%) | 0.46 |
| Race | |||
| White | 574 (76%) | 19 (63%) | 0.04 |
| Black | 71 (9%) | 7 (23%) | |
| Asian | 25 (3%) | 2 (7%) | |
| Other | 91 (12%) | 2 (7%) | |
| Known genetic syndrome | 76 (10%) | 3 (10%) | 1.0 |
| Payer | |||
| Private insurance | 450 (59%) | 17 (57%) | 0.34 |
| Medicaid | 207 (27%) | 6 (20%) | |
| Other | 112 (15%) | 7 (23%) | |
| Mortality | 7 (0.9%) | 0 | 1.0 |
| Cost (2012US$) ∗ | 43,877 (range: 4,399–866,720, IQR: 35,294 – 57,045) | 47,609 (range: 10,532-89,771, IQR: 35,310-72,024) | 0.45 |
| Hospital length of stay (days) | 4 (range: 1-126, IQR: 3-5) | 1 (range: 1-14, IQR: 1-2) | <0.0001 |
| ICU stay (days) † | 2 (range: 1-41, IQR: 2-4) | 2 (range: 1-4, IQR: 1.5-3) | 0.21 |
∗ n=713 for SVR and n=26 for TVR.
The annual rate of PVR increased over the study period. The total rate of PVR across 35 centers increased significantly (β = 12.3 more PVRs each year, 95% confidence interval [CI] 9.8 to 14.7, p <0.001; Figure 1 ). Forty PVRs were performed in 2004, which more than tripled by 2012, when 138 PVRs were performed. The mean rate of PVR by center also increased significantly (β = 0.3 PVRs per year, 95% CI 0.2 to 0.5, p <0.001; Figure 2 ), with an annual rate of 1.1 PVRs per year in 2004 increasing to 3.9 PVR per year in 2012. When adjusted for center total surgical volume, PVRs per center continued to increase without a change in the rate (β = 0.3 PVRs per year, 95% CI 0.19 to 0.40, p <0.001). The proportion of PVRs within each center’s annual cardiac surgical volume also increased significantly (β = 0.7 PVRs per 1,000 cardiac surgical cases per center per year, 95% CI 0.4 to 0.9, p <0.001).
The multivariate mixed-effects model assessing the association between date of PVR and age at PVR is listed in Table 3 . The adjusted expected age at PVR is 16.0 years (95% CI 13.5 to 18.5). There was no association between date of PVR and age of subjects who underwent PVR (β = 0.09, 95% CI −0.1 to 0.3 p = 0.50; Figure 3 ). Several covariates demonstrated significant associations with age at PVR. Black and Asian race (β = −2.9, p = 0.003, and β = −2.1, p = 0.04, respectively) were associated with earlier PVR relative to subjects with white race, as was the presence of genetic syndrome (β −1.8, p = 0.03). Male gender was associated with older age at PVR (β = 1.8, p = 0.003).
| beta | 95% CI | p | |
|---|---|---|---|
| Date of PVR (per year from 2004) | 0.08 | -0.3 to 0.4 | 0.65 |
| Race (compared to white) | |||
| Black | -2.9 | -4.7 to -1.0 | 0.002 |
| Asian | -2.1 | -4.2 to -0.1 | 0.04 |
| Other | -1.0 | -2.7 to 0.6 | 0.23 |
| Male sex (compared to female) | 1.8 | 0.6 to 3.0 | 0.003 |
| Known genetic syndrome | -1.8 | -3.4 to -0.14 | 0.03 |
| Payor (compared to private insurance) | |||
| Medicaid | 0.4 | -1.4 to 2.2 | 0.66 |
| Other | -1.1 | -2.8 to 0.6 | 0.20 |
| Trans-catheter pulmonary valve replacement (compared to operative valve replacement) | 3.9 | -0.7 to 8.5 | 0.10 |
Age at PVR was significantly heterogenous among centers (p <0.001; Figure 4 ), with mean age by center ranging from 13 to 27.8 years. The association between age at PVR and date of PVR did not differ significantly among centers (p = 0.86).
Three sensitivity analyses were performed, as described in the “Methods” section. The first assessed for nonlinear associations between date of PVR and age at PVR, with no improvement in model fit (data not shown). The second assessed the effect of different exclusion criteria on the basis of age by varying them, with no change in the magnitude or direction of main effects or covariates (data not shown). The third excluded subjects who underwent transcatheter PVR, with no change in the magnitude or direction of main effects and covariates (data not shown).
Standardized hospital LOS was 5.4 days (95% CI 4.1 to 7.1; Table 4 ). It was not related to date of operation (coefficient = 0.98, 95% CI 0.95 to 1.0, p = 0.06) or age at operation (coefficient = 1.0, 95% CI 1.0 to 1.0, p = 0.41). Presence of genetic syndrome (coefficient = 1.30, 95% CI 1.09 to 1.60, p = 0.01) and Medicaid (vs private insurance; coefficient = 1.29, 95% CI 1.09 to 1.53, p = 0.003) were associated with longer hospital LOS. TC-PVR was associated with reduced hospital LOS (coefficient = 0.39, 95% CI 0.26 to 0.61, p <0.001). Duration of intensive care was available in 81% of subjects (656 of 799) from all 36 centers. Standardized ICU duration was 2.8 days (95% CI 2.0 to 3.9). Duration of ICU stay did not vary across the study period (95% CI 1.0 to 1.07, p = 0.08; Table 5 ). Genetic syndrome (coefficient = 1.26, 95% CI 1.01 to 1.56, p = 0.04) and older age (coefficient = 1.01 per year greater than 16, 95% CI 1.00 to 1.01, p = 0.03) and Medicaid (vs private insurance; coefficient 1.17, 95% CI 1.01 to 1.36, p = 0.03) were independently associated with increased ICU stay.
