[Skip to Content]
[Skip to Content Landing]
Figure 1.
Study Cohort Selection
Study Cohort Selection

This flow chart shows reasons for record exclusion and points of record linkage.

Figure 2.
Trend in Proportion of Biological and Mechanical Valves Implanted
Trend in Proportion of Biological and Mechanical Valves Implanted

This line graph demonstrates an increase in biological prosthesis implantation rates from 59% to 86% over the study period.

Figure 3.
Trend in Number of Valves Implanted By Valve Series
Trend in Number of Valves Implanted By Valve Series

These graphs illustrate implantation counts by each valve series. For the purpose of this study, Vascutek series also includes Koehler Medical and AorTech valves.

Figure 4.
Patient Age Distributions at Time of Surgery
Patient Age Distributions at Time of Surgery

Patient age is stratified by valve series. The boxes represent quartiles; whiskers represent most extreme data points which are no more than 1.5 times the interquartile range from the box. For the purpose of this study, Vascutek series also includes Koehler Medical and AorTech valves.

Figure 5.
Valve Series for Time to Death or Reintervention
Valve Series for Time to Death or Reintervention

Frailty effects (filled circles) and 95% prediction intervals (black lines) are calculated for Cox random effects models (with and without adjustment for other patient risk factors and operative risk factors). Blue dotted lines indicates there was no effect. For the purposes of this study, Vascutek series also includes Koehler Medical and AorTech valves. To the best of our knowledge, the Ultracor mechanical valve was not acquired by Vascutek, with ownership residing with Koehler.

Supplement.

eAppendix 1. Methods and Data: Further Details

eAppendix 2. Additional Results

eReferences.

eFigure 1. Box-and-whisker plots of logistic EuroSCORE stratified by valve series. In the interests of visual clarity, outliers have been removed

eFigure 2. Distribution of patient gender stratified by valve series

eFigure 3. Distribution of native valve pathology stratified by valve series

eFigure 4. Distribution of preoperative left ventricular ejection fraction (LVEF) stratified by valve series

eFigure 5. Distribution of concomitant CABG surgery stratified by valve series

eFigure 6. Distribution of preoperative body mass index (BMI) stratified by valve series

eFigure 7. Distribution of dyspnoea (New York Heart Association [NYHA] grade) stratified by valve series

eFigure 8. Mean (±1 standard deviation) implanted prosthetic valve size stratified by valve series

eFigure 9. Kaplan-Meier plots of freedom from re-intervention or death by valve series for biological valves

eFigure 10. Kaplan-Meier plots of freedom from re-intervention or death by valve series for mechanical valves

eFigure 11. Frailties (black filled circles) and 95% prediction intervals (black lines) by valve series for time-to-death or re-intervention as calculated for Cox random effects models (with [right panel] and without [left panel] adjustment for other patient and operative risk factors). Red dashed line denotes ‘no effect’. Only biological valves implanted after 1st April 2008 were included in the sensitivity analysis.

eFigure 12. Frailties (black filled circles) and 95% prediction intervals (black lines) by valve series for time-to-death or re-intervention as calculated for Cox random effects models (without adjustment for baseline covariates) with a time origin of 90-days. Patients who died or had a re-intervention on postoperative day 90 or before were excluded, as were patients with <90-days follow-up time. Red dashed line denotes ‘no effect.’

1.
Rathi  VK, Krumholz  HM, Masoudi  FA, Ross  JS.  Characteristics of clinical studies conducted over the total product life cycle of high-risk therapeutic medical devices receiving FDA premarket approval in 2010 and 2011.  JAMA. 2015;314(6):604-612.PubMedGoogle ScholarCrossref
2.
Ross  JS.  Strengthening medical device postmarket safety surveillance.  JAMA Intern Med. 2015;175(8):1350-1351.PubMedGoogle ScholarCrossref
3.
Hauser  RG.  Here we go again--failure of postmarketing device surveillance.  N Engl J Med. 2012;366(10):873-875.PubMedGoogle ScholarCrossref
4.
Maijers  MC, Niessen  FB.  Prevalence of rupture in poly implant Prothèse silicone breast implants, recalled from the European market in 2010.  Plast Reconstr Surg. 2012;129(6):1372-1378.PubMedGoogle ScholarCrossref
5.
Smith  AJ, Dieppe  P, Vernon  K, Porter  M, Blom  AW; National Joint Registry of England and Wales.  Failure rates of stemmed metal-on-metal hip replacements: analysis of data from the National Joint Registry of England and Wales.  Lancet. 2012;379(9822):1199-1204.PubMedGoogle ScholarCrossref
6.
Dalmau  MJ, González-Santos  JM, Blázquez  JA,  et al.  Hemodynamic performance of the Medtronic Mosaic and Perimount Magna aortic bioprostheses: five-year results of a prospectively randomized study.  Eur J Cardiothorac Surg. 2011;39(6):844-852.PubMedGoogle ScholarCrossref
7.
Jamieson  WRE, David  TE, Feindel  CMS, Miyagishima  RT, Germann  E.  Performance of the Carpentier-Edwards SAV and Hancock-II porcine bioprostheses in aortic valve replacement.  J Heart Valve Dis. 2002;11(3):424-430.PubMedGoogle Scholar
8.
David  TE, Armstrong  S, Maganti  M.  Hancock II bioprosthesis for aortic valve replacement: the gold standard of bioprosthetic valves durability?  Ann Thorac Surg. 2010;90(3):775-781.PubMedGoogle ScholarCrossref
9.
Taylor  KM, Gray  SA, Livingstone  S, Brannan  JJ.  The United Kingdom heart valve registry.  J Heart Valve Dis. 1992;1(2):152-159.PubMedGoogle Scholar
10.
Keogh  BE, Kinsman  R.  The Society for Cardiothoracic Surgery in Great Britain & Ireland: The Fifth National Adult Cardiac Surgical Database Report. Henley-on-Thames, UK: Dendrite Clinical Systems Ltd; 2003.
11.
Taylor  K.  The United Kingdom Heart Valve Registry: the first 10 years.  Heart. 1997;77(4):295-296.PubMedGoogle ScholarCrossref
12.
Bridgewater  B; Society for Cardiothoracic Surgery in GB and Ireland.  Cardiac registers: the adult cardiac surgery register.  Heart. 2010;96(18):1441-1443.PubMedGoogle ScholarCrossref
13.
Kumar  A, Matheny  ME, Ho  KKL,  et al.  The data extraction and longitudinal trend analysis network study of distributed automated postmarket cardiovascular device safety surveillance.  Circ Cardiovasc Qual Outcomes. 2015;8(1):38-46.PubMedGoogle ScholarCrossref
14.
Rumsfeld  JS, Peterson  ED.  Achieving meaningful device surveillance: from reaction to proaction.  JAMA. 2010;304(18):2065-2066.PubMedGoogle ScholarCrossref
15.
Hickey  GL, Grant  SW, Cosgriff  R,  et al.  Clinical registries: governance, management, analysis and applications.  Eur J Cardiothorac Surg. 2013;44(4):605-614.PubMedGoogle ScholarCrossref
16.
Hickey  GL, Cosgriff  R, Grant  SW,  et al.  A technical review of the United Kingdom National Adult Cardiac Surgery Governance Analysis 2008-11.  Eur J Cardiothorac Surg. 2014;45(2):225-233.PubMedGoogle ScholarCrossref
17.
Alvarez  JR, Sierra  J, Vega  M,  et al.  Early calcification of the aortic Mitroflow pericardial bioprosthesis in the elderly.  Interact Cardiovasc Thorac Surg. 2009;9(5):842-846.PubMedGoogle ScholarCrossref
18.
Butany  J, Feng  T, Luk  A, Law  K, Suri  R, Nair  V.  Modes of failure in explanted mitroflow pericardial valves.  Ann Thorac Surg. 2011;92(5):1621-1627.PubMedGoogle ScholarCrossref
19.
Yankah  CA, Pasic  M, Musci  M,  et al.  Aortic valve replacement with the Mitroflow pericardial bioprosthesis: durability results up to 21 years.  J Thorac Cardiovasc Surg. 2008;136(3):688-696.PubMedGoogle ScholarCrossref
20.
ISTHMUS Investigators.  The Italian study on the Mitroflow postoperative results (ISTHMUS): a 20-year, multicentre evaluation of Mitroflow pericardial bioprosthesis.  Eur J Cardiothorac Surg. 2011;39(1):18-26.Google ScholarCrossref
21.
Krucoff  MW, Sedrakyan  A, Normand  S-LT.  Bridging unmet medical device ecosystem needs with strategically coordinated registries networks.  JAMA. 2015;314(16):1691-1692.PubMedGoogle ScholarCrossref
22.
Rising  J, Moscovitch  B.  The Food and Drug Administration’s unique device identification system: better postmarket data on the safety and effectiveness of medical devices.  JAMA Intern Med. 2014;174(11):1719-1720.PubMedGoogle ScholarCrossref
23.
Vidi  VD, Matheny  ME, Donnelly  S, Resnic  FS.  An evaluation of a distributed medical device safety surveillance system: the DELTA network study.  Contemp Clin Trials. 2011;32(3):309-317.PubMedGoogle ScholarCrossref
24.
Resnic  FS, Gross  TP, Marinac-Dabic  D,  et al.  Automated surveillance to detect postprocedure safety signals of approved cardiovascular devices.  JAMA. 2010;304(18):2019-2027.PubMedGoogle ScholarCrossref
25.
Moore  TJ, Furberg  CD.  Electronic health data for postmarket surveillance: a vision not realized.  Drug Saf. 2015;38(7):601-610.PubMedGoogle ScholarCrossref
26.
Bridgewater  B, Hickey  GL, Cooper  G, Deanfield  J, Roxburgh  J; Society for Cardiothoracic Surgery in Great Britain and Ireland; National Institute for Clinical Outcomes Research, UCL.  Publishing cardiac surgery mortality rates: lessons for other specialties.  BMJ. 2013;346:f1139.PubMedGoogle ScholarCrossref
27.
Shahian  DM, Edwards  FH, Jacobs  JP,  et al.  Public reporting of cardiac surgery performance: Part 2--implementation.  Ann Thorac Surg. 2011;92(3)(suppl):S12-S23.PubMedGoogle ScholarCrossref
28.
Hickey  GL, Grant  SW, Murphy  GJ,  et al.  Dynamic trends in cardiac surgery: why the logistic EuroSCORE is no longer suitable for contemporary cardiac surgery and implications for future risk models.  Eur J Cardiothorac Surg. 2013;43(6):1146-1152.PubMedGoogle ScholarCrossref
29.
Horton  NJ, Kleinman  KP.  Much ado about nothing: a comparison of missing data methods and software to fit incomplete data regression models.  Am Stat. 2007;61(1):79-90.PubMedGoogle ScholarCrossref
30.
Hernán  MA.  The hazards of hazard ratios.  Epidemiology. 2010;21(1):13-15.PubMedGoogle ScholarCrossref
31.
Grunkemeier  GL, Jin  R, Eijkemans  MJC, Takkenberg  JJM.  Actual and actuarial probabilities of competing risks: apples and lemons.  Ann Thorac Surg. 2007;83(5):1586-1592.PubMedGoogle ScholarCrossref
32.
Akins  CW, Miller  DC, Turina  MI,  et al.  Guidelines for reporting mortality and morbidity after cardiac valve interventions.  Eur J Cardiothorac Surg. 2008;33(4):523-528.PubMedGoogle ScholarCrossref
33.
Taylor  KM.  Acute failure of artificial heart valves.  BMJ. 1988;297(6655):996-997.PubMedGoogle ScholarCrossref
Original Investigation
January 2017

National Registry Data and Record Linkage to Inform Postmarket Surveillance of Prosthetic Aortic Valve Models Over 15 Years

Author Affiliations
  • 1University of Liverpool, Department of Biostatistics, Liverpool, L69 3GL, England
  • 2University College London, National Institute for Cardiovascular Outcomes Research (NICOR), London, EC1A 4NP, England
  • 3Computer Science Corporation, Kings Cross, London, N1C 4AG, England
  • 4Academic Surgery Unit, University of Manchester, Manchester Academic Health Science Centre, University Hospital of South Manchester, Manchester, M23 9LT, UK
  • 5Department of Cardiac Surgery, Bristol Heart Institute, Bristol Royal Infirmary, Bristol, BS2 8HW, England
  • 6South West Cardiothoracic Centre, Derriford Hospital, Derriford, Plymouth, PL68DH, England
  • 7Cardiovascular Biomedical Research Unit, Royal Brompton Hospital, London, SW3 6NP, England
  • 8University of Manchester, Manchester Academic Health Science Centre, Centre for Health Informatics, Vaughan House, Manchester, M13 9GB, England
  • 9Department of Cardiothoracic Surgery, James Cook University Hospital, Middlesbrough, TS4 3BW, England
 

Copyright 2016 American Medical Association. All Rights Reserved.

JAMA Intern Med. 2017;177(1):79-86. doi:10.1001/jamainternmed.2016.6936
Key Points

Question  Do prosthetic aortic valves implanted in patients operated on in England and Wales display unexpected patterns of reintervention-free survival?

Findings  Using a UK national clinical registry, records for more than 54 000 aortic valve replacement procedures were mapped to implanted prosthetic valves; 2 biological valves displayed patterns of increased hazard, and 3 displayed decreased hazard.

Meaning  In the absence of alternative data sources, clinical registry data should be used to inform postmarket surveillance of cardiovascular devices.

Abstract

Importance  Postmarket evidence generation for medical devices is important yet limited for prosthetic aortic valve devices in the United Kingdom.

Objective  To identify prosthetic aortic valve models that display unexpected patterns of mortality or reintervention using routinely collected national registry data and record linkage.

Design, Setting, and Participants  This observational study used data from all National Health Service and private hospitals in England and Wales that submit data to the National Adult Cardiac Surgery Audit (NACSA). All patients undergoing first-time elective and urgent aortic valve replacement surgery (with or without coronary artery bypass grafting) with a biological (n = 15 series) or mechanical (n = 10 series) prosthetic valve from 5 primary suppliers, and satisfying prespecified data quality criteria (n = 43 782 biological; n = 11 084 mechanical) between 1998 and 2013 were included. Valves were classified into series of related models. Outcome tracking was performed using multifaceted record linkage. The median follow-up was 4.1 years (maximum, 15.3 years). Cox proportional hazards regression with random effects (frailty models) were used to model valve effects on the outcomes, with and without adjustment for preoperative and intraoperative covariates.

Main Outcomes and Measures  Time to all-cause mortality or aortic valve reintervention (surgical or transcatheter). There were 13 104 deaths and 723 reinterventions during follow-up.

Results  Of 79 345 isolated aortic valve replacement procedures with or without coronary artery bypass grafting, 54 866 were analyzed. Biological valve implantation rates increased from 59% in 1998 and 1999 to 86% in 2012 and 2013. Two series of valves associated with significantly increased hazard of death or reintervention were identified (first series: frailty, 1.18; 95% prediction interval [PI], 1.06-1.32 and second series: frailty, 1.19; 95% PI, 1.09-1.31). These results were robust to covariate adjustment and sensitivity analyses. There were 3 prosthetic valves with a significant reduction in hazard (valve 1: frailty, 0.88; 95% PI, 0.80-0.96; valve 2: frailty, 0.88; 95% PI, 0.80-0.96; and valve 3: frailty, 0.88; 95% PI, 0.78-0.98).

Conclusions and Relevance  Meaningful evidence from the analysis of routinely collected registry data can inform postmarket surveillance of medical devices. Although the findings are associated with a number of caveats, 2 specific biological aortic valve series identified in this study may warrant further investigation.

Introduction

In recent years there has been a shift in emphasis from establishing device safety and effectiveness before marketing to postmarket evidence generation and surveillence.1 The US Food and Drug Administration (FDA) system for postmarket surveillance has been found to be in need of strengthening.2,3 This need for improved surveillance systems was recently highlighted by the international health scare caused by Poly Implant Prothèse breast implants.4 Postmarket surveillance systems have historically been reactive rather than proactive in the United Kingdom, as evidenced by concerns over hip prostheses leading to the UK National Joint Registry being established.5

Prosthetic heart valves have evolved significantly since the first valve replacement was performed in 1952. Although there are 2 main groups of prosthetic heart valve—tissue or mechanical—there are a variety of different valves within these groups. There is a large body of literature on the long-term reliability of prosthetic heart valves, but these studies, whether randomized trials,6 observational comparison studies,7 or single case series,8 typically compare a very small number of valves. Data from systematic benchmarking of long-term performance is not readily available.

In the United Kingdom, the Heart Valve Registry was established in 1986 between the Government Department of Health and the Society for Cardiothoracic Surgery in Great Britain and Ireland.9,10 Within 10 years a minimum data set of clinical variables about heart valve replacement procedures had been entered for more than 45 000 patients.11 The United Kingdom Heart Valve Registry fulfilled an important role: the ability to monitor trends in outcomes by different prosthetic valve models. It was setup to do this by recording valve model and serial numbers for implanted prosthetic valves and also by linkage to mortality data, including cause of death, from the Office for National Statistics. In 2004, funding was withdrawn owing to cost and governance issues, with its functionality partly subsumed by a national adult cardiac surgery register.12 Currently, the UK agency responsible for ensuring that medical devices meet applicable standards of safety—the Medicines and Healthcare products Regulatory Agency (MHRA)—collects data on acute valve failures submitted by health care professionals; however, in the absence of a device-specific registry, the opportunity to detect patterns of unexpected outcomes is limited.

Prospective surveillance based on clinical registries that record device-specific information can identify important signals that passive reporting mechanisms may miss,13,14 and there have been calls to move from reactive to proactive monitoring.14 As a prelude to any prospective surveillance program, we present results for a retrospective cross-sectional analysis of prosthetic valves implanted into patients undergoing aortic valve replacement (AVR) surgery with or without concomitant coronary artery bypass grafting in England and Wales over the past 15 years.

Methods
Extraction and Preprocessing of Aortic Valve Surgery Data

A complete extract from the National Adult Cardiac Surgery Audit (NACSA; registry version 4.1.2), which is run by the National Institute for Cardiovascular Outcomes Research (NICOR; an institute of University College London), was performed on October 10, 2014. This extract included all adult cardiac surgery procedures performed in UK National Health Service (NHS) hospitals, some private hospitals and some hospitals in the Republic of Ireland. Case ascertainment of NHS procedure is expected to be high for most of the study period.12 As part of a wider clinical epidemiological research and quality improvement program, a regularly updated suite of “data cleaning” rules developed by specialist clinicians were coded and applied to the raw data (excluding the valve model data) prior to any analysis as summarized in eAppendix 1 in the Supplement.15,16 This study was approved by the NICOR NACSA Research Board, and the need to obtain informed consent from patients was waived as patient identifiable information was either removed or pseudonymized.

The initial filtering step was to extract all records corresponding to aortic valve surgery performed in hospitals located in England and Wales between April 1, 1998, and March 31, 2013. Data for 1 private hospital were removed prior to analysis pending local validation, as were all data for patients who had more than 1 record in the registry for the same admission spell. For the purposes of this study, we selected all patients who underwent an AVR with or without coronary artery bypass grafting. We then excluded all records corresponding to: (1) patients having previous cardiac surgery; (2) suspected incorrectly entered transcatheter aortic valve implantation (TAVI) procedures (as identified using a rules-based approach); (3) emergency or salvage procedures; (4) unidentifiable responsible consultant surgeon (as identified by a unique surgeon’s General Medical Council number in the registry); (5) missing primary outcome data.

Record Linkage

To facilitate long-term monitoring of patient and valve status, we performed multiple record linkages for each patient for life status, surgical reoperation, and TAVI as described in eAppendix 1 in the Supplement.

Valve Model Data and Data Quality

Prostheses are recorded in the NACSA registry in 2 separate free-text fields: valve name and valve model. There was inconsistency on how each hospital entered these data. An updated suite of data-processing scripts was written to map each recorded name and/or model to a homogenous list of known prosthetic valves using a variety of information sources as described in eAppendix 2 in the Supplement. For each record, we attempted to record the valve manufacturer, model, series, and type (mechanical or biological, and xenograft type in the case of biological valves). Here, “series” refers to a group of valve models from a single manufacturer considered related (eTable in the Supplement). Not all valves could be accurately classified. Note that manufacturer classification only reflects association as of 2015 to the best of our knowledge. Some models have been acquired by manufacturers through business mergers and acquisitions, but are grouped together according to model.

Records that were irrelevant or featured gross inconsistencies were excluded, including records that could either not be matched or which were matched to more than one manufacturer, series, or type, or that were matched to more than 1 model were excluded. Homografts, autografts, rings, valve conduits, 2 particular model types, off-label procedures, and valves not produced by one of the UK primary suppliers were also excluded (eAppendix 2 in the Supplement). The 5 manufacturers included are Edwards Lifesciences, Medtronic Inc, Sorin Group, St Jude Medical Inc, and Vascutek Ltd. For the purposes of this study, we also include Koehler Medical and AorTech valves in the Vascutek group owing to changes in ownership of some models between these companies over the study period.

Study Variables

For each procedure, data were extracted for administrative factors, patient characteristics, comorbidities, surgical team, intraoperative factors, and postoperative outcomes. There were few missing clinical data (all >95% complete with the exception of the dichotomous creatinine variable [5.6% missing], critical preoperative state [7.4% missing], hemodynamics [5.1% missing], and aortic valve pathology [10.1% missing]). Details of study variable definitions and missing data imputation are given in eAppendix 2 in the Supplement.

Study Outcomes

The outcome for this study was time from surgery to the first event of death or reintervention. Patients were censored at the last follow-up time if alive and reintervention free. Patients who died in hospital on the day of surgery were recorded as having a nominal survival time of 0.5 days. Follow-up data, until the point of discharge, were collected by the NACSA registry, and postdischarge survival data were collected by record linkage to the Office for National Statistics death registry. Reintervention was defined as surgical reoperation on the aortic valve for any reason or TAVI. Time-to-reintervention data was collected by intrarecord and interrecord linkage as described above.

Statistical Analysis

Mechanical and biological valves were analyzed separately to avoid confounding by indication. Valves were compared only at series level. The Kaplan-Meier estimator was used to construct survival curves for the time-to-event outcome and compared between valves using log-rank tests. Multivariable Cox proportional hazards regression models were used to adjust for potential differences with zero-mean valve series-level normally distributed random effects. The exponentiated random-effects—also known as the shared frailties—act multiplicatively on the baseline hazard rate and therefore have an intuitive translation: frailty terms greater than 1 correspond to increased hazard for a valve, and those less than 1 correspond to decreased hazard. Frailties where the corresponding 95% prediction interval lower limit lies above 1 indicate a valve with a significantly large hazard rate for the outcome. The focus of this study was not the identification of prognostic factors, hence we limit reporting to the frailty effects. For comparison, unadjusted frailties are also reported. All analyses and data cleaning were performed in R version 3.3.1 (R Foundation for Statistical Computing; http://www.R-project.org/). A more detailed description of the statistical analysis is given in eAppendix 1 in the Supplement. A number of different sensitivity analyses were performed (eAppendix 1 in the Supplement). All inferences remained broadly consistent.

Results

From 79 345 AVR with or without coronary artery bypass grafting records with a biological or mechanical prosthesis, a total of 54 866 records were retained for analysis (Figure 1; eAppendix 2 in the Supplement), from 37 hospitals (including 4 private units) and 344 consultant surgeons.

eTable 1 in the Supplement lists the valves included, which were grouped into 15 and 10 series of biological and mechanical valves, respectively. Figure 2 shows an increasing trend in the implantation rate of biological valves during the study period, stabilizing at 86%. Figure 3 shows the number of valves implanted by time for each series. The distribution of patient age at surgery (Figure 4) indicates homogeneity between the valve series within type (biological and mechanical), with the exception of greater patient ages for the Medtronic Hall series, Vascutek (including Koehler Medical and AorTech) Ultracor series, Edwards Lifesciences mechanical series, and Sorin Sutureless series relative to others of the same type. Plots for logistic EuroSCORE, sex, native valve pathology, procedure, left ventricular ejection fraction, body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), valve size, and New York Heart Association grade are shown in eFigures 1 through 8 in the Supplement.

Valve Outcomes

During a median follow-up of 4.1 years (maximum follow-up 15.3 years), 13 104 deaths (11 353 biological; 1751 mechanical) were recorded and there were 723 reinterventions (571 biological; 152 mechanical; 682 were surgical procedures and 41 were TAVIs). Results from the Kaplan-Meier estimator analysis and pathological data for surgical reinterventions are described in the eAppendix in the Supplement.

After adjustment, the random effects survival model indicated that the Sorin Mitroflow series (frailty, 1.19; 95% prediction interval [PI], 1.09-1.31]) and Sorin Biological series (frailty, 1.18; 95% PI, 1.06-1.32) displayed larger hazard than expected (Figure 5). To place the outcomes into perspective, the 10-year overall freedom from reintervention or death rates for the 2 valves were 33.8% (95% CI, 31.3%-36.5%) and 41.4% (95% CI, 37.6%-45.6%), respectively, compared with the overall average of 47.2% (95% CI, 46.2%-48.1%) for all non-Sorin biological valves. Although nonsignificant, the lower 95% PI for the Medtronic ATS-3f series only marginally crossed the line of unity (frailty, 1.21; 95% PI, 1.00-1.47]). For mechanical valves, the Medtronic Hall valve had a significantly larger unadjusted hazard (unadjusted frailty, 1.48; 95% PI, 1.22-1.80). However, after adjustment this was diminished (adjusted frailty, 1.10; 95% PI, 0.97-1.24), reflecting the greater patient age relative to the profile of other mechanical valves. Additional results are provided in the eAppendix in the Supplement.

There were 3 prosthetic valves with a significant reduction in hazard (Figure 5): the Edwards Lifesciences Perimount series (frailty, 0.88; 95% PI, 0.80-0.96]), the Edwards Lifesciences Perimount Magna series (frailty, 0.88; 95% PI, 0.80-0.96), and the Medtronic Hancock series (frailty, 0.88; 95% PI, 0.78-0.98).

A subgroup analysis of all bioprosthesis records performed on or after April 1, 2008, (n = 23 834) showed that the lower 95% PI limit was less than 1 for every valve after adjustment (eAppendix in the Supplement).

Discussion

We analyzed a comprehensive clinical registry to measure reintervention-free survival in a large series of patients undergoing AVR in the United Kingdom. Two series of prosthetic aortic valves were associated with significantly increased hazards of death or reintervention, relative to the population of prosthetic valves implanted in England and Wales from large suppliers. Similarly, 3 series of prosthetic valves were associated with decreased hazards. Inferences remained broadly consistent following covariate adjustment and sensitivity analyses. This study has shown that routinely collected clinical registry data can be exploited, in conjunction with multifaceted record linkage, to perform long-term device surveillance.

There is a large literature examining outcomes following different prosthetic AVR implants. Few studies, however, reflect national data. Moreover, the evidence base is mixed. For example, some studies have suggested an inferior performance of the Sorin Mitroflow17,18 whereas others have demonstrated long-term durability and hemodynamic performance.19,20

The National Health Service number—a unique patient identifier—enables record linkage across clinical registries and other data sources. It would be feasible to exploit this to link across further data sources (eg, trace readmission from administrative data). In fact, strategic linking of complementary registries and data sources is a “foundational architectural construct” recommendation of the US Medical Device Registries Task Force.21 Furthermore, record linkage could be further extended using unique serial numbers of implanted devices (including prosthetic aortic valves) to device manufacturer databases to improve ongoing research, augment clinical trial follow-up after completion, and to allow traceability in case of serious fault detection. The planned role out by the FDA of a unique device identification system integrated for use with electronic health records would allow scalable cross-specialty surveillance.22

We explored outcomes in prosthetic valves cross-sectionally using 15 years of data. Moving forward, this is not a suitable approach for postmarket device surveillance, which should be dynamic, providing regular updates, to achieve superiority over existing passive reporting mechanisms. It is conceivable that signals of unexpected patterns of outcomes could have been detected earlier on. The Data Extraction and Longitudinal Trend Analysis (DELTA) network study23 is a validated example of such a tool, which has used propensity score matching and statistical process control methodology to evaluate the safety of high-risk cardiovascular devices for perioperative binary outcomes.13,21,24 Similar efforts for postmarket surveillance of pharmacological products are also ongoing.25 While the methodology applied here was relatively simplistic, what we have demonstrated is that routinely collected clinical registry data can be leveraged for evaluating performance of medical devices, even when this was not a primary goal of the data collection program. With some improvements to the data collection mechanisms, this messy real-world registry, or other registries, data could be analyzed using alternative platforms.

Limitations
Data Quality

Research with routinely collected health care data inevitably raises questions over data quality. Many of the data on clinical variables are of high quality, owing to the fact they are used for national governance.16,26 Valve-specific data, on the other hand, are not subject to similar quality management. Because the valve model data were collected as free-text inputs, more data quality issues were present than for equivalent clinical information collected using structured inputs. Data quality is expected to improve in the future, due to increased scrutiny of device monitoring. Caution must therefore be taken when interpreting the results, because there is potential for coding errors by the surgeon.

Valve Classification

Focusing surveillance on a coarsened valve grouping (ie, series) as opposed to valve models ensured that the maximum number of records would be available for analysis. This decision, while allowing us to retain more records for analysis, introduces limitations. First, different models in a series, including stented and stentless models, or different generations of the same model, might have a variable effect on outcome. For example, the latest generations of Sorin Mitroflow valves are processed with a phospholipid reduction treatment to mitigate calcification. This might lead to improved performance compared with earlier generations. Second, not all valve series are clearly delineated owing to either historical device company purchases and/or mergers or naming conventions. Similarity in naming means that valves identified to the series level but not the model level might potentially be misclassified. This is discussed further in the eAppendix in the Supplement.

Covariate Adjustment

The adjustment data used in this study derives from a national clinical registry, which is widely accepted to be superior to administrative data.27 There was no a priori expectation of gross selection bias by valve series within valve type, nor was substantial heterogeneity observed, unlike in some other postmarket surveillance studies for cardiovascular devices.13 However, there has been a shift in patient risk profiles over time,28 which might confound with market availability of certain valves. We adjusted for baseline risk factors, as well for a number of clinical valve-related variables, and contrasted the change in inference with that of the unadjusted model. Another potential source of bias stems from the missing data being imputed according to a (sex-stratified) mean or mode approach;29 however, missing data was not considered substantial. It should also be noted that the number of random effects was quite small for a frailty model. Additionally, no adjustment for institutional effects were included, which could conflate with models implanted.

Study Outcomes

In some records, patient identification was missing, which can reduce the ability to track patients. Moreover, tracking was terminated at different time points for different end points: December 2012 for TAVI, March 2013 for surgical reintervention, and July 2013 for survival. Because the focus of the study was on valve surveillance rather than patient outcomes monitoring, we only analyzed the time to first event. We also note that sample sizes differed substantially between models. This was owing to multiple factors, including market availability; some are relatively new and others have been withdrawn, which might affect the ability to detect valves that have significantly different event hazards.30 Some newer implanted valves may not yet have sufficient volume to show significantly different outcomes. Differences in outcomes may be attributable to different causes; however, we have defined a composite outcome for analysis, rather than analyzing death and reintervention as a competing risk.31

The greatest clinical limitations of this study are its relatively short follow-up and lack of other clinical outcomes.32 The median follow-up time was 4.1 years; however, valve failure is most likely to occur later on, especially in the context of mechanical valves. In fact, only 152 surgical reinterventions were observed in the mechanical valve group. Finally, we excluded patients who had multiple surgical records within a single admission; however, there were only 34 such cases satisfying this exclusion criteria for the study.

Conclusions

The need for such postmarketing surveillance of medical devices was made clear by the Poly Implant Prothèse breast implant and other medical device scares,3 yet infrastructure is lacking. We have shown that a national clinical registry, linked to other routinely collected data, might be used to inform postmarket surveillance programs. By analyzing 15 years of data on AVR procedures in England and Wales, we identified 2 prosthetic valves that may warrant further scrutiny through additional studies. As Taylor noted about valve monitoring nearly 3-decades ago, “overreaction is as inappropriate as complacency.”33 Given the limitations of the study, the signals shown here should only serve as hypothesis generating and not misinterpreted as causal effects.

Back to top
Article Information

Corresponding Author: Joel Dunning PhD, FRCS (C-Th), Department of Cardiothoracic Surgery, James Cook University Hospital, Middlesbrough, TS4 3BW, UK (joeldunning@doctors.org.uk).

Correction: This article was corrected for an incorrect reference on November 28, 2016.

Published Online: November 7, 2016. doi:10.1001/jamainternmed.2016.6936

Author Contributions: Dr Hickey had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Hickey, Bridgewater, Grant, Moat, Buchan, Dunning.

Acquisition, analysis, or interpretation of data: Hickey, Grant, Deanfield, Parkinson, Bryan, Dalrymple-Hay, Dunning.

Drafting of the manuscript: Hickey, Bridgewater, Grant, Buchan, Dunning.

Critical revision of the manuscript for important intellectual content: All Authors.

Statistical analysis: Hickey, Grant, Dunning.

Obtained funding: Bridgewater, Deanfield, Buchan.

Administrative, technical, or material support: Bridgewater, Deanfield, Parkinson.

Conflict of Interest Disclosures: Dr Bridgewater has received honoraria from Edwards Lifesciences in the last 3-years. Dr Dunning has received proctoring fees from Cardica. Dr Stuart Grant has received travel and conference fees from Edwards Lifesciences. Dr Dalrymple-Hay has received honoraria from St Jude Medical, Edwards Lifesciences and Maquet, and research grants from Edwards Lifesciences and St. Jude Edwards. Dr Moat has received fees for consultancy and proctoring from Medtronic, consultancy for Tendyne (Direct Flow), and speaking from Abbott. No other conflicts are reported.

Funding/Support: Dr Hickey is currently supported by the Medical Research Council (grant MR/M013227/1). Data preprocessing was completed when Dr Hickey was at the University of Manchester, supported by Heart Research UK (grant RG2583). Dr Buchan is supported by the Medical Research Council funded Health eResearch Centre (grant MR/K006665/1). Dr Deanfield is supported by the British Heart Foundation (grant CH/03/002/15570).

Role of the Funder/Sponsor: The funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Information: The United Kingdom National Adult Cardiac Surgery Audit registry and the United Kingdom National TAVI Audit Registry are available to researchers upon application to the National Institute of Cardiovascular Outcomes Research (NICOR), University College London. Full details on the NICOR data sharing application process are available at https://www.ucl.ac.uk/nicor/access/application (accessed October 16, 2015).

Additional Contributions: The authors acknowledge all members of the Society for Cardiothoracic Surgery in Great Britain and Ireland who contribute data to the Society for Cardiothoracic Surgery in Great Britain and Ireland database. The National UK TAVI Registry steering group agreed to the use of record linkage for the purposes of tracking valve-in-valve re-interventions. The NICOR, University College London, provided the data for this study. The authors also thank Miss Rebecca Cosgriff (former National Adult Cardiac Surgery Audit project manager for NICOR) for obtaining the hospital code lists; Mr John McKenna (formerly of Vascutek Terumo) for helping to identify some of the valves; and representatives of the valve companies and hospital database managers for their helpful support. No one acknowledged here received or provided compensation for their services.

References
1.
Rathi  VK, Krumholz  HM, Masoudi  FA, Ross  JS.  Characteristics of clinical studies conducted over the total product life cycle of high-risk therapeutic medical devices receiving FDA premarket approval in 2010 and 2011.  JAMA. 2015;314(6):604-612.PubMedGoogle ScholarCrossref
2.
Ross  JS.  Strengthening medical device postmarket safety surveillance.  JAMA Intern Med. 2015;175(8):1350-1351.PubMedGoogle ScholarCrossref
3.
Hauser  RG.  Here we go again--failure of postmarketing device surveillance.  N Engl J Med. 2012;366(10):873-875.PubMedGoogle ScholarCrossref
4.
Maijers  MC, Niessen  FB.  Prevalence of rupture in poly implant Prothèse silicone breast implants, recalled from the European market in 2010.  Plast Reconstr Surg. 2012;129(6):1372-1378.PubMedGoogle ScholarCrossref
5.
Smith  AJ, Dieppe  P, Vernon  K, Porter  M, Blom  AW; National Joint Registry of England and Wales.  Failure rates of stemmed metal-on-metal hip replacements: analysis of data from the National Joint Registry of England and Wales.  Lancet. 2012;379(9822):1199-1204.PubMedGoogle ScholarCrossref
6.
Dalmau  MJ, González-Santos  JM, Blázquez  JA,  et al.  Hemodynamic performance of the Medtronic Mosaic and Perimount Magna aortic bioprostheses: five-year results of a prospectively randomized study.  Eur J Cardiothorac Surg. 2011;39(6):844-852.PubMedGoogle ScholarCrossref
7.
Jamieson  WRE, David  TE, Feindel  CMS, Miyagishima  RT, Germann  E.  Performance of the Carpentier-Edwards SAV and Hancock-II porcine bioprostheses in aortic valve replacement.  J Heart Valve Dis. 2002;11(3):424-430.PubMedGoogle Scholar
8.
David  TE, Armstrong  S, Maganti  M.  Hancock II bioprosthesis for aortic valve replacement: the gold standard of bioprosthetic valves durability?  Ann Thorac Surg. 2010;90(3):775-781.PubMedGoogle ScholarCrossref
9.
Taylor  KM, Gray  SA, Livingstone  S, Brannan  JJ.  The United Kingdom heart valve registry.  J Heart Valve Dis. 1992;1(2):152-159.PubMedGoogle Scholar
10.
Keogh  BE, Kinsman  R.  The Society for Cardiothoracic Surgery in Great Britain & Ireland: The Fifth National Adult Cardiac Surgical Database Report. Henley-on-Thames, UK: Dendrite Clinical Systems Ltd; 2003.
11.
Taylor  K.  The United Kingdom Heart Valve Registry: the first 10 years.  Heart. 1997;77(4):295-296.PubMedGoogle ScholarCrossref
12.
Bridgewater  B; Society for Cardiothoracic Surgery in GB and Ireland.  Cardiac registers: the adult cardiac surgery register.  Heart. 2010;96(18):1441-1443.PubMedGoogle ScholarCrossref
13.
Kumar  A, Matheny  ME, Ho  KKL,  et al.  The data extraction and longitudinal trend analysis network study of distributed automated postmarket cardiovascular device safety surveillance.  Circ Cardiovasc Qual Outcomes. 2015;8(1):38-46.PubMedGoogle ScholarCrossref
14.
Rumsfeld  JS, Peterson  ED.  Achieving meaningful device surveillance: from reaction to proaction.  JAMA. 2010;304(18):2065-2066.PubMedGoogle ScholarCrossref
15.
Hickey  GL, Grant  SW, Cosgriff  R,  et al.  Clinical registries: governance, management, analysis and applications.  Eur J Cardiothorac Surg. 2013;44(4):605-614.PubMedGoogle ScholarCrossref
16.
Hickey  GL, Cosgriff  R, Grant  SW,  et al.  A technical review of the United Kingdom National Adult Cardiac Surgery Governance Analysis 2008-11.  Eur J Cardiothorac Surg. 2014;45(2):225-233.PubMedGoogle ScholarCrossref
17.
Alvarez  JR, Sierra  J, Vega  M,  et al.  Early calcification of the aortic Mitroflow pericardial bioprosthesis in the elderly.  Interact Cardiovasc Thorac Surg. 2009;9(5):842-846.PubMedGoogle ScholarCrossref
18.
Butany  J, Feng  T, Luk  A, Law  K, Suri  R, Nair  V.  Modes of failure in explanted mitroflow pericardial valves.  Ann Thorac Surg. 2011;92(5):1621-1627.PubMedGoogle ScholarCrossref
19.
Yankah  CA, Pasic  M, Musci  M,  et al.  Aortic valve replacement with the Mitroflow pericardial bioprosthesis: durability results up to 21 years.  J Thorac Cardiovasc Surg. 2008;136(3):688-696.PubMedGoogle ScholarCrossref
20.
ISTHMUS Investigators.  The Italian study on the Mitroflow postoperative results (ISTHMUS): a 20-year, multicentre evaluation of Mitroflow pericardial bioprosthesis.  Eur J Cardiothorac Surg. 2011;39(1):18-26.Google ScholarCrossref
21.
Krucoff  MW, Sedrakyan  A, Normand  S-LT.  Bridging unmet medical device ecosystem needs with strategically coordinated registries networks.  JAMA. 2015;314(16):1691-1692.PubMedGoogle ScholarCrossref
22.
Rising  J, Moscovitch  B.  The Food and Drug Administration’s unique device identification system: better postmarket data on the safety and effectiveness of medical devices.  JAMA Intern Med. 2014;174(11):1719-1720.PubMedGoogle ScholarCrossref
23.
Vidi  VD, Matheny  ME, Donnelly  S, Resnic  FS.  An evaluation of a distributed medical device safety surveillance system: the DELTA network study.  Contemp Clin Trials. 2011;32(3):309-317.PubMedGoogle ScholarCrossref
24.
Resnic  FS, Gross  TP, Marinac-Dabic  D,  et al.  Automated surveillance to detect postprocedure safety signals of approved cardiovascular devices.  JAMA. 2010;304(18):2019-2027.PubMedGoogle ScholarCrossref
25.
Moore  TJ, Furberg  CD.  Electronic health data for postmarket surveillance: a vision not realized.  Drug Saf. 2015;38(7):601-610.PubMedGoogle ScholarCrossref
26.
Bridgewater  B, Hickey  GL, Cooper  G, Deanfield  J, Roxburgh  J; Society for Cardiothoracic Surgery in Great Britain and Ireland; National Institute for Clinical Outcomes Research, UCL.  Publishing cardiac surgery mortality rates: lessons for other specialties.  BMJ. 2013;346:f1139.PubMedGoogle ScholarCrossref
27.
Shahian  DM, Edwards  FH, Jacobs  JP,  et al.  Public reporting of cardiac surgery performance: Part 2--implementation.  Ann Thorac Surg. 2011;92(3)(suppl):S12-S23.PubMedGoogle ScholarCrossref
28.
Hickey  GL, Grant  SW, Murphy  GJ,  et al.  Dynamic trends in cardiac surgery: why the logistic EuroSCORE is no longer suitable for contemporary cardiac surgery and implications for future risk models.  Eur J Cardiothorac Surg. 2013;43(6):1146-1152.PubMedGoogle ScholarCrossref
29.
Horton  NJ, Kleinman  KP.  Much ado about nothing: a comparison of missing data methods and software to fit incomplete data regression models.  Am Stat. 2007;61(1):79-90.PubMedGoogle ScholarCrossref
30.
Hernán  MA.  The hazards of hazard ratios.  Epidemiology. 2010;21(1):13-15.PubMedGoogle ScholarCrossref
31.
Grunkemeier  GL, Jin  R, Eijkemans  MJC, Takkenberg  JJM.  Actual and actuarial probabilities of competing risks: apples and lemons.  Ann Thorac Surg. 2007;83(5):1586-1592.PubMedGoogle ScholarCrossref
32.
Akins  CW, Miller  DC, Turina  MI,  et al.  Guidelines for reporting mortality and morbidity after cardiac valve interventions.  Eur J Cardiothorac Surg. 2008;33(4):523-528.PubMedGoogle ScholarCrossref
33.
Taylor  KM.  Acute failure of artificial heart valves.  BMJ. 1988;297(6655):996-997.PubMedGoogle ScholarCrossref
×