Patient-reported outcomes (PROs) such as pain and functional ability, perceptions of physical and mental health, and quality of life in a global sense, are increasingly used to assess the impact of joint replacement surgery [1–4]. PROs can be employed to guide clinical care decisions, monitor quality of care, perform between-hospital comparisons [5], or adjust reimbursement policies [6].

PROs are often understood to be linked causally; e.g., specific symptoms such as pain or impairment determine global perceptions of health, which in turn contribute to perceived quality of life [7]. If the relationships between symptoms and more global PROs were stable, it would be easy to predict the impact of symptom relief, such as can be afforded by joint replacement surgery, on general health. However, relationships between PROs may vary with context. E.g., the same level of impairment may be perceived differently if the problem is temporary rather than permanent, if the patient develops other health issues, or if the patient reevaluates his or her priorities [8–10]. In patients with degenerative joint disease, a key contextual variable is the occurrence of joint replacement surgery. Before surgery, some patients may consider osteoarthritis symptoms as temporary and curable, and thus more tolerable. It is currently unclear if the impact of pain or functional impairment on perceived health remains stable or changes after joint replacement surgery. This issue is important for the interpretation of changes in PROs following surgery.

In this study, we examined the stability of the relationship between patients’ symptoms (pain and functional impairment) and global perceptions of physical and mental health, by comparing the periods before and after hip or knee replacement surgery. We do not focus on the absolute changes in PROs brought on by surgery, but rather on the impact of surgery on the associations between PROs.

During the study period, 1969 THAs and 1834 TKAs were performed at the hospital (Fig 1), and 990 THAs in 935 patients and 907 TKAs in 842 patients completed both questionnaires and were included in this analysis.

The differences between arthroplasties that were included in the analysis and those which were excluded due to unavailable questionnaire data were generally small, with few exceptions. In THA patients (we use the term somewhat loosely here, as the unit of observation is the THA, even when two THAs were performed in the same patient), the non-participants were on average younger (67.0 years vs 68.6, p = 0.009), more likely to smoke (24.5% vs 17.9%, p<0.001) and more likely to have an ASA score of 3 (21.0% vs 15.9%, p = 0.004); differences were small and statistically non-significant in terms of sex, body mass index, and prevalence of heart disease, diabetes, and high blood pressure. In TKA patients, the non-participants were also younger (70.5 years vs 71.5, p = 0.025), had a higher body mass index (30.3 vs 29.1, p = 0.018), were more likely to be women (69.8% vs 65.5%, p = 0.048), to smoke (14.4% vs 10.7%, p = 0.018) and to have an ASA score of 3 (24.4% vs 17.3%, p<0.001); differences were small and non-significant as to comorbid conditions.

Among study participants, TKA patients were older than THA patients (mean 71.5 years (SD 9.4) vs 68.6 (11.9)), they were also more likely to be female, obese, non-smokers, and more likely to suffer from diabetes or high blood pressure (Table 1). Pre-operative pain and activity levels were similar in the THA and TKA cohorts.

This prospective study showed a substantial change in the relationships between symptoms (i.e., pain and function) and patients’ perceptions of mental health following total joint arthroplasty. Mental health scores were influenced to the same extent by pain and function after surgery as before (same slopes in the linear regression model), but their absolute level was lowered after surgery at a given level of pain and function (lower intercepts). For physical health, the absolute scores at mid-range values of pain and function were unchanged after surgery (same intercepts), but the associations became slightly stronger (steeper slopes). These results suggest that relationships between PROs are context-dependent, and that one should be careful when interpreting changes in PROs over time, especially if the changes are induced by surgery.

Total joint arthroplasty delivered very substantial pain relief and functional improvement, and increased physical health scores by more than a standard deviation on average. The improvement was less impressive for mental health (2–3 points on MCS scale), corroborating previous reports [19]. Yet, after surgery, perceived mental health was substantially lower than would have been predicted from the improvements in pain and function, as if perceived mental health was recalibrated downward. The apparent reduction in mental health was of about a half standard deviation (5 points), which can be considered as clinically meaningful [20, 21]. On the other hand, the slope of the association, which represents the sensitivity of perceived mental health to the intensity of symptoms, remained of the same magnitude as before surgery, as did the variance of mental health explained by the WOMAC Pain or Function scales. Thus pain and function remained important determinants of mental health after surgery, only with the absolute level of perceived mental health shifted downward. In other words, higher levels of function and pain relief were required after surgery than before to achieve a given level of mental health.

We observed no such systematic shift for post-operative perceptions of physical health. For an average patient, the improvement in PCS was exactly of the amount that could be predicted from the cross-sectional association observed before surgery, as though change in pain and function fully explained the improvement in perceived physical health. Furthermore, the slope of the association and the proportion of explained variance in physical health increased somewhat after surgery. This may reflect a patient’s greater focus on symptoms of osteoarthritis during postoperative rehabilitation.

The relationships beween symptom scales and measures of mental and physical health changed in a similar way for patients with both types of joint replacement, THA and TKA. This suggests that these phenomena may have general validity. It would be interesting to see if other types of interventions that have a notable impact on chronic symptoms and functional limitations (such as correction of sight or hearing, asthma medication, prosthetic limbs, or restoration of coronary flow), induce similar shifts in associations between PROs.

We can propose several possible explanations for our findings. The downward shift in MCS scores may be attributed to unfulfilled expectations. Pain and functional impairment may be less tolerable to some patients after total joint arthroplasty than before, because they may have expected better results from a surgical intervention that is often understood to be of last resort [22]. Other factors, such as a residual limp after surgery, may also have a negative psychological impact on patients [23]. Another possibility is that the surgery itself, the hospitalization and the subsequent rehabilitation were emotionally taxing, and this may have negated in part the beneficial effects of improved pain and function. Furthermore it is possible that mental health is inherently more stable than physical health, as if it reflected constant personality traits [24], and therefore is less amenable to change. This notion is supported by our observation of higher intraclass correlation coefficients for the MCS than for the PCS.

Of note, recent studies have shown a similar phenomenon: the presence of severe pain and dysfunction decreased the global perception of health (measured on a visual analog scale between 0 and 100) more strongly after total hip [25] or knee [26] arthroplasty than before. These studies used the EuroQol EQ5D instrument, which does not separate global perceptions of mental and physical health, and therefore do not replicate our findings. They do demonstrate, however, that associations between symptoms and health perceptions can be context-dependent, specifically in patients with total joint arhtroplasty.

Finally, it is possible that the measurement properties of the instruments changed following surgery and that what we have observed are variants of information bias [8]. Previous studies have described a “recalibration” to lower values of PRO instruments following total knee arthroplasty [27, 28], but the phenomenon was similar for symptom scales and for health status scales. In our study, we have observed changes in associations between symptom scales and health perceptions, which makes information bias less likely. In another analysis of the same patient cohorts, we observed that the self-rated health item remained completely unchanged after total joint arthroplasty [10], which provides further evidence that the measurement properties of psychometric instruments vary with context.

A strong feature of our study is the inclusion of large unselected cohorts of patients, which yields precise estimates and grants real world relevance to our results. As the same patients were assessed at baseline and at follow-up, confounding by stable patient characteristics cannot explain the observed changes.

The main limitation of our study is the lack of a clear mechanistic explanation for the observed results, which cannot be inferred from observed data. Qualitative interviews with patients before and after joint replacement surgery may improve our understanding of the contextual determinants of patients’ reports, and of the mechanisms that caused the observed changes in relationships between psychometric PRO scales. Nevertheless, the consistency of the findings in two cohorts of patients indicates that the phenomena are real. Another limitation is that we were unable to include in this analysis all possible determinants of the MCS and PCS, such as psychiatric and somatic comorbidities. However, to the extent that such determinants remain stable over the 1-year period, they would not have influenced the comparison of pre-operative and post-operative associations between WOMAC scores and SF12 scores. Finally, the study participants represented about one half of all eligible patients, mostly due to failure to return the self-report questionnaires. Study participants were somewhat older but healthier than the non-participants. However, we believe that it is unlikely that this selection process would have modified the patterns of association between health status and symptom scales, which is the main theme of this analysis.

Our results may not influence clinical decisions for individual patients, but they are relevant for the measurement of PROs as indicators of the quality of care provided by surgeons or hospitals. Clinical outcomes were not influenced in the same way even by a highly effective intervention, and their relationships were context-dependent. This suggests that measurement of multiple PROs, some disease-specific and others generic, is to be preferred in order to capture the complexity of each patient’s experience. Our results also illustrate a novel manifestation of response shift, one that affects relationships between PROs.

We found that perceptions of mental health were lower after total joint arthroplasty of the hip and knee than may have been expected based on symptom relief, whereas the perceptions of physical health were as predicted by symptom relief. The reasons for these phenomena should be explored, in order to facilitate the interpretation of post-intervention PRO scores.