Currently most registries provide detailed information on implants and surgery, but have little information on patient characteristics and outcomes other than implant revision. Registries must collect this additional information, which will then permit investigators (1) to analyse additional indicators of success and failure other than revision (e.g. PROMs) as well as surrogates of failure (e.g. early abnormal radiographic findings [
38]), (2) to adjust for potential confounders when comparing treatments, (3) to evaluate causal mechanisms, and (4) to develop a personalised approach to treatment. Thus, registries should capture patient co-morbidities and health behaviours such as smoking and obesity, which could confound or modify the risks of adverse events [
39]. Although the most robust and complete data need to be gathered prospectively as part of the primary data collection of the joint registries, reduced data completeness and accuracy may jeopardize this goal. In practice there is currently greater reliance on linkage to secondary data sources to obtain additional data. Secondary sources include primary and secondary care data and data from registries such as for mortality, cancer or drugs. For example, researchers in the United Kingdom recently linked the National Joint Registry to the Clinical Practice Research Datalink to study safety issues related to the use of metal-on-metal implants [
40], and also to the Hospital Episode Statistics (inpatient records) to compare uni-compartmental versus total knee replacement [
41]. In the latter case, the more detailed information about the patient’s characteristics at the time of surgery and about reoperations and readmissions - obtained from the inpatient records - increased the number of outcomes evaluated as well as the ability to adjust for differences between the two treatment groups. Another example is the linkage to databases that record medications (e.g. prolonged antibiotics or pain medication use), which has been shown to offer a useful surrogate measure for prosthesis infection [
42] and revision [
43]. In addition, linking to a system designed for spontaneously reporting adverse events [
44] may have the potential to improve the detection of failures. Finally, integrating health economics data within registries via primary data collection and/or linkage to secondary data may improve clinical and public health decision-making.
Improving data analysis requires speeding up failure detection and bias minimisation including, but not limited to, confounding by indication. First, stakeholders including manufacturers need to adopt strategies to improve post-authorisation safety surveillance, which are increasingly used to detect adverse events in vaccine and drug surveillance [
45,
46] and have become routine when assessing drug post-marketing (EU Regulation No 1027/2012). Regulators should prioritise real-time monitoring of devices by analysing specific risks [
47]. Secondly, researchers, regulators and manufacturers should systematically use measures of benefit and risk [
48,
49] including PROMs of pain, function and activity, health-related quality of life, satisfaction, and costs to compare devices from a societal and policy-makers’ perspective. Third, when possible randomized trials should be nested in registries. This has the potential to combine the advantages of both study designs and to facilitate the conduct of multi-centre trials with reduced duration and costs [
50,
51]. Fourth, researchers should test and incorporate methods (e.g. propensity score methods, sequential cohort analyses among others) developed to reduce bias and confounding when evaluating drugs and vaccines in observational studies [
52‐
55]. Fifth, there is a need to stratify the risk of implant failure and other adverse events by factors specific to patients, surgery, implant, and environment. This may allow stakeholders to target improvements to subgroups, and to inform case-mix adjustment models. Finally, methods for data analyses at an aggregate level should be applied to estimate the comparative effectiveness of multiple treatments [
56].