Reliability of Breast Implants

Dear Larry

Thank you for your data request for breast implant data and apologies for the delay in responding. The data available is:

• The number of women receiving implants, by year, by major manufacturer
• Number of Explants: All Manufacturers (inc. Others and Unknown Brands)

My colleagues have been copied into this email to show your request has been actioned. I hope this is helpful.

Kind regards

Amina.., Department of Health, Quarry House, UK

Amina’s email went on to say…”On data quality you might like to note that: two patients apparently received PIP implants after their recall. This was most likely a data entry error. Since errors of this kind in a very small number of data entries would have no material impact on the primary purpose of the analysis-i.e. to compare the performance of PIP and other implants-we did not investigate further. Analysing the explant data by month would involve disproportionate cost and would not add materially to the accuracy of the overall analysis.”

Background

UK National Health Service (NHS) did breast implants, some elective and some reconstructive, starting in 2001. In the mid-2000s, leakage and rupture was cause for some explants. Patients sued to get UK NHS to pay for those explants. Manufacturer “Poly Implant Prosthèse” (PIP) (France) was implicated. PIP claimed quality control problem and had swapped industrial for medical grade silicone. A UK “Expert Group” investigated, concurred, and paid [Keough]. I asked for data, and Amina…, Department of Health, sent annual implant and explant counts for nonparametric survival analyses (tables 1 and 2).

Table 1. Implant data by manufacturer (M1, M2, M3, and PIP) 

Table 2. Explant data by manufacturer

Survival Analyses

The FDA and others require serial numbers on implantable medical and arthroplasty devices [FDA, Ranstam et al.] devices, so that times to failures or censoring times can be collected. It is unnecessary to track implant patients by name or their implants by serial number to provide early warning, management by exception, estimate survival functions, and forecast explants. Implant and explant counts are statistically sufficient to make populationnonparametric estimates of what biostatisticians call survival functions and what engineers call reliability functions: P[Time from implant to explant>t], where t = 1, 2,…., are calendar time periods. These estimates show differences between manufacturers and changes within manufacturers’ products over time, without tracking implants by patient name and implant serial number. 

Figure 1 shows the nonparametric reliability function estimates by manufacturer. Most explants occur within the first year after implant (~5% varying by manufacturer). Manufacturer M2 probably had problems (lowest reliability, orange line, “M2 nplse”). US implants in early 2000s, except for Allergan (AbbVie), were made in Brazil: Mentor (J&J Santa Barbara and Irving, TX), Sientra-Silimed (Garland, TX), and a few others. PIP shows an additional ~0.3-0.4% occurring four or five years after implant. Manufacturer M2 shows an additional almost 0.1% occurring between six and nine years after implant. The BMI-PIP rupture reliability estimate shows ~2% ruptures beginning in the fourth year after implant. “PIP BMI Rupture” is based on life data from one source BMI (highest reliability, red line). BMI is a British HealthCare company, not “Body-Mass-Index”.

Table 3. Probability of explant within first year after implant

Figure 2 includes BMI-PIP rupture survival function estimate obtained from the expert group report. It is the highest reliability estimate, because rupture is a subset of PIP explant modes. 

Figure 2 shows BMI-PIP rupture explant broom chart. Survival function estimates for broom charts were computed using successively smaller, earlier subsets of cohort data. The broom chart shows survival function estimates from subsets from: 2001-2002, 2001-2003, 2001-2004,…,all. Early 2001-2006 PIP implants were the cohorts with rupture problems. This is an example of reliability deterioration, then growth. Despite the early problems, the broom chart shows that implant survival function eventually improved; longer lines include longer implant-explant lives as more years are included in the estimates. The first-year PIP explant probability estimate 0. 46% did not change. 

Methods

The methods were nonparametric maximum likelihood (npmle) and least squares (nplse) survival function estimation [George]. The PIP npmle and nplse survival function estimates agreed closely. The broom chart survival function estimates give the standard deviations of the estimates, at each age t =1,2,…,5. 

Table 4 shows the  lower confidence limit (LCL) on average survival function estimates from all manufacturers’ data (“All LCL”), and from PIP data (“PIP LCL”}. PIP LCLs are slightly greater than all LCLs indicating that PIP survival function could be better than average despite the BMI-PIP rupture data. That conclusions is not statistically significant, because of the slight variations between manufacturers, Table 4 is not a “confidence band” for all ages shown [Hall and Wellner]. Table 4 gives indications only. 

Table 4. Lower confidence limits on survival function estimates, average survival function minus one standard deviation. 

Proportions of implants for reconstruction varied between manufacturers. PIP market share for reconstruction was smallest and dwindled to zero over time since 2001. Other factors such as hospital, patient, and experience may affect results. Please let me know if you have questions, would like the Excel workbook, or would like more information or additional computations, such as population nonparametric reliability function estimates by failure mode, without life data.

References

Docket FDA-2012-N-0359, “Strengthening our National System for Medical Device Postmarket Surveillance”  Sept. 2012

L. L. George, “Estimate Reliability Functions Without Life Data,” ASQ Reliability Review, Vol. 13, pp. 21-25, 1993

…, “Actuarial Forecasts, Least Squares Reliability, and Martingales,” https://fred-schenkelberg-project.prev01.rmkr.net/actuarial-forecasts-least-squares-reliability-martingales/#more-421021, 2022

W. J. Hall and Jon A. Wellner , “Confidence Bands for a Survival Curve from Censored Data,” Biometrika, Vol. 67, No. 1, pp. 133-143, April 1980

Jonas Ranstam, Johan Kärrholm, Pekka Pulkkinen, Keijo Mäkelä, Birgitte Espehaug, Alma Becic Pedersen, Frank Mehnert, and Ove Furnes, “Statistical Analysis of Arthroplasty Data,” Acta Orthopaedica, 82 (3), pp. 253-257, 2011

Sir Bruce Keough, “Poly Implant Prosthèse, (PIP) breast implants: Final report of the Expert Group,”https://www.gov.uk/government/publications/poly-implant-prothese-pip-breast-implants-final-report-of-the-expert-group, June 2012

… Volume 2, Appendices, June 2012