Abstract
Background:
The use of registry data to detect and eliminate inferior devices is based on the assumption that the results of the first cases performed with a new device are indicative of how the same implant would perform with widespread usage. However, existing registry data clearly proves that the performance of individual implants is very surgeon dependent. In this study we utilized a computer simulation of a large implant registry to address the question: How does the pairing of different surgeons with different implants affect the ability of registries to correctly identify inferior devices?
Materials and Methods:
A synthetic implant registry was created consisting of 10,000 patients who underwent joint replacement performed by 100 different surgeons using 5 different implants. Hazard functions representing the relative risks for revision associated with individual patients and surgeons were derived from the annual reports of implant registries. The cumulative revision rates (CRR values) of the 5 hypothetical implants were fixed at nominal values of 10%, 15%, 20%, 25%, and 30% at 15 years post operation vs. 10% for average implants. The surgeons were ordered according to their individual probabilities of a revision at less than 15 years post-op. Each surgeon was placed in one of 8 subsets comprised of 12.5% of the total surgeon pool, ranging from the lowest to the highest risk of revision. Patients, surgeons, and implants were randomly matched in an iterative fashion to simulate 500 separate RCTs, starting with the group of surgeons of with the lowest risk, and then repeating the simulation using surgeons with the lowest and second lowest risk of revision. This process was repeated iteratively until all surgeons were enrolled.
Results:
The performance of each implant on the registry simulations was highly dependent on the composition of the surgeon pool. When surgeon participation was limited to the upper half of the pool (ie those with above-average survivorship), all five implant designs had mean CRRs less than the average device (9.1% vs 10%). Furthermore, if the 25% of surgeons with the highest failure rates did not use the worst performing of the 5 designs, its mean revision rate dropped below the detection threshold of the registry (200% of average). The impact of surgeon pairing varied with the failure rate of each implant. Overall, the bottom 25% of surgeons performed 52% of cases that went to failure, however this percentage ranged from 73% of cases performed with the average implant to 40% with the 3x design.
Conclusions:
-
1.
Identification of “poorly-performing” implants using registry data is highly susceptible to errors due to non-random pairing of surgeons and devices.
-
2.
If highly skilled and experienced surgeons perform initial series using new devices, there is a high probability that inferior performance will be masked and that even more patients will be placed at risk of premature failure.
-
3.
It is critical that registries devise methods to adjust survivorship data for variations in surgeon participation before extrapolating the outcome of early cases to generalized usage of new devices.
