Quantifying Physical Resilience after Knee or Hip Surgery in Older Australian Women Based on Long-Term Physical Functioning Trajectories

Knee surgery and hip surgery are typically done with the aim of improving function and reducing pain. However, particularly in older adults, there is great variation in recovery after surgery. A systematic review of 14 papers describing 6 hip and 11 knee cohorts found that 7–23% of hip surgery patients and 10–34% of knee surgery patients reported chronic pain post-surgery, with follow-up ranging from 3 months to 5 years post-surgery [1]. In a cohort of patients with end-stage knee osteoarthritis, 72% improved in function after knee replacement surgery, but 25% did not improve and 4% experienced further decline up to 7 years post-surgery [2]. In a cohort of 626 osteoarthritis patients, 14% of hip patients and 20% of knee patients did not meaningfully improve in pain and functioning up to 5 years post-replacement surgery [3]. A better understanding of variation in recovery trajectories and its causes provides valuable information for shared decision-making and personalising care pre- and post-surgery to optimise recovery [4]. For example, prehabilitation has been suggested to optimise fitness before surgery with the aim to improve recovery after surgery. Systematic reviews have provided inconclusive evidence for the effectiveness of prehabilitation [5, 6]. A potential way to increase its cost-effectiveness may be by offering prehabilitation only to patients with the highest risk of poor recovery [7]. However, that would require an accurate tool to identify patients with poor recovery in the pre-surgery stage, which is not currently available.

A resilience approach can help understand variation in recovery capacity [4]. Physical resilience can be defined as the individual’s capacity to remain well, recover, or even thrive in response to a health stressor, such as surgery [8, 9]. Various methods to quantify physical resilience have been described [10, 11]. The choice of method depends on the aim and context in which it is applied [9, 10, 12]. One study quantified physical resilience in response to hip fracture using a statistical approach based on the difference between predicted and observed outcomes [11], also referred to as the residuals approach [10]. In a cohort of 541 hip fracture patients, they found that 12.5% of patients recovered better than expected, whereas in 13.4% of patients, recovery was lower than expected up to 12 months post-fracture [11]. Another study used an a priori approach and operationalised “high resilience” as absence of delirium in response to elective knee or hip surgery [13]. In that study, younger age, higher intelligence scores, a lower preoperative pain score on movement, and a lower concentration of T-tau were independently associated with resistance to delirium up to 3 days post-surgery.

Population-based studies of ageing showed that the average trajectory of physical functioning (PF) follows a gradual, non-linear decline that becomes more rapid with advanced ages [14]. When examining the long-term trajectories of functioning after knee or hip surgery, the ageing-related decline in PF should be taken into account. Few studies to date have examined the recovery in PF after knee or hip surgery from a resilience perspective. The studies that have been done were limited to 12 months of follow-up after surgery [11] or used different outcomes [13].

The first aim was to compare trajectories of PF over time in older Australian women with and without hip or knee surgery, in the context of ageing-related decline. With “trajectory,” we refer to the course in PF over time. A priori, we expected to observe a more rapid decline pre-surgery and recovery or stabilisation after surgery in the women with surgery compared with the women without surgery. The second aim was to quantify physical resilience in response to knee and hip surgery in a population-based cohort of older Australian women. To quantify resilience, we used the expected recovery potential method [11], also referred to as the residuals approach [10]. This is a statistics-driven method that accounts for age- and comorbidity-related decline [11]. Profiles of women with higher and lower resilience were described to provide insight into who tends to have better or worse recovery capacity.

This observational study uses data from participants in the Australian Longitudinal Study on Women’s Health. This is an ongoing, population-based study on the health and well-being of four generations of women. Recruitment and data collection procedures have been described elsewhere [15]. Briefly, women were randomly selected from the national Medicare health insurance database, which includes all citizens and permanent residents of Australia. Apart from overrepresentation of Australian-born and higher-educated women, the sample was representative of the general population of Australian women [15]. The study protocol was reviewed and approved by the University of Newcastle Human Research Ethics Committee (Approval No. H-076-0795) and The University of Queensland Human Research Ethics Committee (Approval No. 2004/HE000224). All participants provided signed informed consent.

For the current analyses, we used data from the oldest of the four cohorts. These women were born in 1921–26. A total of 12,432 women completed the baseline postal mail survey in 1996. Follow-up surveys were completed at 3-year intervals until 2011. Data were included from women who returned survey 2 (1999) and could be classified as having or not having had knee or hip surgery at surveys 2 to 6 (1999–2011) (online suppl. Fig. 1; for all online suppl. material, see https://doi.org/10.1159/000540159). As survey 1 did not include questions about knee or hip surgery, data from this survey were not included in the current analysis.

PF was measured as the PF subscale of the Medical Outcomes Survey 36-Item Short-Form Health Survey (SF-36) at each survey [16‒18]. The subscale score for PF was calculated in accordance with the SF-36 manual [19]. The score ranges from 0 to 100 with higher scores indicating better functioning.

In survey 2, surgery was measured with the question “Have you had any of the following operations or procedures?”. If participants answered “yes, in the past 3 years” to the response options “hip surgery” or “knee surgery or arthroscopy,” they were classified as having had hip or knee surgery, respectively. In surveys 3–6, surgery was measured with the question “In the last 3 years, have you had any of the following operations or procedures?”. If participants answered “yes” to the response options “hip surgery or hip replacement” or “knee surgery or arthroscopy,” they were classified as having had hip or knee surgery, respectively. Responses to these questions in surveys 2–6 were combined into one variable for hip surgery and one variable for knee surgery, indicating whether they reported surgery at any of these surveys (yes vs. no).

Sociodemographic (age, area of residence, marital status, and community dwelling) characteristics were measured using self-reports at every survey and categorised as listed in Table 1. Level of education was measured in survey 1 (1996) as the highest qualification completed: “no formal qualification,” “school certificate,” “high school certificate,” “trade/apprenticeship/diploma,” “university or higher degree.” Health variables were also measured using self-reports in each survey. BMI was calculated based on self-reported height and weight. Number of chronic conditions was measured as the sum of doctor-diagnosed or treated conditions in the past 3 years, including arthritis, diabetes, cardiovascular disease, hypertension, stroke, lung disease, osteoporosis, cancer, depression, or dementia (range 0–10). Bodily pain was measured as that domain score of the SF-36 (range 0–100), with higher scores indicating less pain. Joint symptoms and dizziness were measured with the question “Have you had any of the following problems in the last 12 months?”: “stiff or painful joints” and “dizziness, loss of balance.” Response options were collapsed into “often” versus “never/rarely/sometimes.” Falls were measured as “In the last 12 months, have you had a fall to the ground? (does not include stumbles/trips)” (yes/no). Needing help with daily tasks was measured as “Do you regularly need help with daily tasks because of long-term illness, disability or frailty (e.g., personal care, getting around, and preparing meals)?” (yes/no). Physical activity was measured using the modified Active Australia questionnaire [20]. Minutes in the last week spent walking and in leisure-time activities were multiplied by a metabolic equivalent (MET) score (i.e., walking = 3, moderate activities = 4.5, and vigorous activities = 7 [21]) and then summed and categorised as inactive (0–40 MET.min/week), low (40–599 MET.min/week), moderate (600–1,200 MET.min/week), and high (≥1,200 MET.min/week). Reasons for attrition were derived from the participant status file, which differentiates between withdrawn, deceased, and did not complete the survey (due to frailty or could not be contacted).

Descriptive statistics were used to compare the sample characteristics at study baseline (survey 2, 1999) between participants with and without knee surgery and between participants with and without hip surgery. Two-sided tests were conducted, with the alpha set at 0.05. All analyses were done in Stata (StataCorp LLC, version 16.0).

Linear mixed models were used to examine trajectories of PF over time in women with and without surgery. Models were run separately for knee and hip surgery. Time was anchored on survey of first reported surgery. For example, if knee surgery was first reported at survey 3, then survey 3 was labelled as time = 0, survey 2 as time = −3, and surveys 4, 5, and 6 as +3, +6, and +9, respectively. This resulted in a time scale ranging from −12 to +12 years. Participants without surgery at surveys 2–6 were matched with participants with knee or hip surgery on the pattern of survey return, age, and number of chronic conditions. Participants without surgery were assigned the same timing of surveys as the participants with surgery in the same matching group. Models included surgery, time (as a factor variable), the surgery-by-time interaction, and the confounders (measured at each wave). All variables listed in Table 1 were considered potential confounders. Potential confounders were selected based on the literature and included age, education, marital status, housing situation, diabetes, cardiovascular disease, number of chronic conditions, BMI, depression, memory problems, dizziness, falls, vision impairments, and physical activity. Confounders were added only if they were associated with the exposure and outcome and changed the regression coefficient by more than 10% on a forward selection basis. Only the variables age at baseline, level of education, BMI, number of chronic conditions, dizziness, and physical activity met these criteria and were included as confounders. To check the accuracy of this selection process, a sensitivity analysis was run that included all potential confounders. Age and BMI were standardised on the mean. Models were run with a random intercept. Assumptions for linear regression were tested and met. The population mean estimates of the mixed models were used to calculate the mean PF values at each time point within each group to plot the trajectories over time. Additional models were run to test if the slope post-surgery differed from the slope pre-surgery in participants who had surgery. This was done by including a dummy variable indicating pre-survey (0) and post-survey (1), the interaction of that dummy variable with time, and the confounders. The regression coefficient for that interaction term indicates the difference in slope before and after surgery (Δslope).

To quantify physical resilience, we used the expected recovery differential method [11]. This method calculates the difference between the predicted and observed values for PF. The predicted values were derived from the linear mixed model described above. The differences between predicted and observed values were calculated for the time points 0, 3, 6, 9, and 12 years after first reported surgery. The continuous difference scores were dichotomised to distinguish participants with lower (difference <0) and higher (difference >0) resilience. Two resilience variables were created. First, participants were classified as having lower or higher resilience at the time of first reported surgery (T = 0) to reflect short-term recovery. Second, participants were classified as having lower and higher resilience based on the sum of differences over the period 0–12 years after first reported surgery to reflect long-term recovery. To examine how characteristics differed between groups of higher and lower resilience, sample characteristics were compared using descriptive statistics.

To examine the influence of dropout on the main findings, we examined differences between participants who completed the sixth survey versus those who dropped out during follow-up. Also, the main analyses were repeated in the subgroup of participants with complete data on all surveys.

In total, data from 9,099 to 9,115 participants were included in the models for hip and knee surgery, respectively (online suppl. Fig. 1). At survey 2 in 1999, participants who reported surgery during follow-up were slightly older, had more chronic conditions, perceived more bodily pain, and were more likely to report arthritis, joint pain/stiffness, a fall, needing help with daily tasks, being physically inactive, and to die during follow-up (p < 0.05, Table 1). In total, 4,003 participants (38.4% of those who completed survey 2 in 1999) died during follow-up. Of those who did not die, 2,067 dropped out of the study for other reasons (37.1% had withdrawn; 29.1% were too frail; 2.5% could not be contacted). Compared with participants who were still participating in survey 6 in 2011, participants who dropped out were older, had lower levels of education, had more chronic conditions, had more bodily pain, and were more likely to report dizziness, needing help with daily tasks, and being physically inactive (p < 0.01, online suppl. Table 1). However, absolute differences between the groups were small for most characteristics except hip and knee surgery. Participants who were still in the study at survey 6 were more likely to report hip or knee surgery during follow-up (p < 0.001).

In the hip surgery group, there was a more rapid decline in PF in the 6 years prior to surgery and a slower decline in the 6 years after surgery than in the no surgery group (Fig. 1; Table 2). Indeed, the interaction between hip surgery and time in the association with PF was statistically significant (p < 0.001). The difference in slope pre- and post-surgery was statistically significant, and the positive regression coefficient suggested that the slope post-surgery was less steep than the slope pre-surgery (Δslope = 0.7, p < 0.001). Similar patterns over time were found in sensitivity analyses including all potential confounders (online suppl. Fig. 2). Similar results were found in the subsample analyses including participants with complete data (online suppl. Fig. 3; Table 2).

At the time of first reported surgery, participants classified as having higher resilience (n = 422) had higher levels of education (p = 0.02), were more likely to be community dwelling (p = 0.002), were less likely to need help with daily tasks (p < 0.001), and were physically inactive (p < 0.001) than participants with lower resilience (n = 478) (Table 3). Moreover, they had fewer chronic conditions and were less likely to report joint symptoms, pain, dizziness, or a fall (p < 0.001). Participants with more resilient trajectories up to 12 years post-surgery also had higher levels of education and were more likely to report health problems than women with less resilient trajectories (Table 3).

PF was lower in the knee surgery than the no surgery group at all time points; however, there was little difference in rate of decline (Fig. 1; Table 2). The interaction between knee surgery and time in the association with PF was statistically significant (p = 0.01). The difference in slope pre- and post-surgery was not statistically significant (Δslope = −0.3, p = 0.25), suggesting no change in rate of decline after surgery. Similar patterns over time were found in sensitivity analyses including all potential confounders (online suppl. Fig. 2). Similar results were found in the subsample analyses including participants with complete data (online suppl. Fig. 3; Table 2).

At the time of first reported surgery, participants classified as having higher resilience (n = 447) were slightly younger (p < 0.001), less likely to need help with daily tasks (p < 0.001), and less likely to be physically inactive (p < 0.001) than participants with lower resilience (n = 432) (Table 4). Moreover, they had fewer chronic conditions, lower BMI, and were less likely to report joint symptoms or bodily pain (p < 0.001). Participants with more resilient trajectories up to 12 years post-surgery were slightly younger, had fewer chronic conditions and symptoms, had lower BMI, less often needed help with daily tasks, and were less likely to be physically inactive than participants with less resilient trajectories (p < 0.006).

In this study, we examined the trajectories of PF from up to 12 years before to 12 years after hip or knee surgery in older Australian women. In women with hip surgery, the decline in PF before surgery was more rapid than in women who did not have hip surgery. From 3 years after hip surgery onwards, the rate of decline in PF was similar in women with and without hip surgery. Women who had knee surgery had lower PF at all times than women who did not have knee surgery. Women with higher resilience to hip or knee surgery had fewer health problems and were more likely to be independent and active than women with lower resilience to hip or knee surgery.

Theoretical trajectories of (high) resilience are typically depicted as a dip in functioning when the stressor occurs, followed by a gradual increase in reverting back to the pre-stressor level of functioning (Fig. 2, left panel) [11, 12]. In this conceptualisation, physical resilience is better if a person’s functioning is less impacted by the stressor, or if the return to pre-stressor level is more complete or faster. Note that sometimes, even an improvement in functioning is desired beyond pre-stressor levels, for example, in the situation of elective knee replacement, which is done with the aim to improve function and reducing pain. In the context of ageing-related decline and long-term follow-up, this theoretical graph should be placed within the context of gradual decline in functioning (Fig. 2, right panel) [14]. In the trajectory for hip surgery, as expected, a clear dip in PF was observed around the time of hip surgery (Fig. 1). In contrast, such a dip in PF was not observed around the time of knee surgery. A few explanations are possible. First, hip surgery is likely a stronger stressor than knee surgery, resulting in a greater dip in PF and longer time to recover. Our definition of knee surgery also included knee arthroscopies, which has less of an impact on PF and from which patients recover more quickly than from, for example, knee replacement surgery. Including knee arthroscopies may have attenuated the potential association between knee surgeries and PF. Given the phrasing of the questionnaires with which these data were collected, we were unable to differentiate between the types of surgery. Second, the 3 years between surveys may have been too long to pick up the decline in functioning and recovery around the time of surgery, as most recovery is expected within 6 months after surgery [22]. Indeed, a study in patients undergoing knee replacement surgery with repeated measures from pre-surgery to 1, 3, 6, and 12 months post-surgery did show such a dip in functional measures around the time of surgery [23]. Third, it could be that other factors have a stronger influence on PF than surgery and that these factors are more prevalent in women with knee surgery than in women with hip surgery. However, Table 2 shows that the characteristics of these two groups are very similar, with the main difference being a higher prevalence of arthritis and joint symptoms in the knee surgery group.

From 3 years after surgery, both hip and knee surgery groups showed a gradual decline in PF. This finding in a population-based study is in line with findings from clinical samples of hip or knee replacement patients, which showed a gradual decline in daily activities, such as self-care and household chores, from 1 to 10 years post-surgery [24, 25]. Our findings add that the rate of decline in PF from 3 years after surgery is similar to that in the general female population of similar age, although on average, the level of functioning remains somewhat lower (Fig. 1; Table 2).

Our findings suggest that women who recovered better than expected after knee or hip surgery had better health and independence than women who recovered worse than expected (Tables 3, 4). One could argue that these women had higher resilience to the surgery. An alternate explanation is that these women already had better functioning before surgery, as the residual approach ignores the pre-surgery level of functioning. Indeed, a study of 541 hip fracture patients undergoing hip replacement found that better self-rated health and daily functioning pre-surgery predicted better recovery at 2, 6, and 12 months post-surgery [26]. Another study of 286 knee replacement patients showed that 7% of patients initially recovered in the first year post-surgery, but then declined in functioning during 1–8 years post-surgery [22]. These patients had worse function and higher BMI pre-surgery and reported more pain post-surgery than the 93% of patients with sustained recovery [22]. Thus, pre-surgery health and functioning are important indicators of post-surgery recovery and long-term functioning.

As explained in the introduction, there are multiple approaches to quantifying physical resilience. In this study, we used the residual approach, which identifies patients with a better-than-average recovery as having the highest resilience [10, 11]. One study compared this residual approach with the “recovery phenotype,” which identifies patients with the best recovery across multiple outcomes as those with the highest resilience [11]. Comparison of the two methods in a sample of 186 acute respiratory infection patients (mean age 70 ± 8.5 years, 48.9% male) showed that each approach identified different patients as more or less physically resilient [11]. Each approach highlights different aspects of resilience, and there may be value in using different approaches in parallel [10, 27]. Our initial plan was to also apply the “a priori” approach, which identifies patients with high resilience based on an a priori defined criterion. Criteria that we considered were greater than the minimal clinically important change in PF or a greater than 1 standard deviation improvement in PF relative to the pre-surgery level of PF. However, these definitions did not take into account the background ageing-related decline in functioning and were therefore considered less appropriate when looking at long-term trajectories, and left out of the paper.

The large, population-based sample with long-term follow-up has both strengths and limitations. On the one hand, the study design is a strength, because it ensures the sample is representative of the general older female population in Australia and provides good reference data for women without surgery. Moreover, data are available for up to 12 years before surgery. This is specifically important for resilience approaches that require a pre-stressor measurement of the outcome. Pre-surgery data are typically not (or limited) available in clinical cohorts of specific patient groups. On the other hand, the long intervals between surveys mean a lower resolution to pick up the changes in functioning immediately before and after the surgery. Also, no information is available on indication for and type of surgery. By looking at any knee or hip surgery, we were unable to specify trajectories for different types of surgeries and underlying pathologies. Thus, data from clinical and population-based samples are complimentary. As described above, findings presented here are in line with those presented in clinical studies.

In this study, we were specifically interested in PF and used that as the sole outcome. Knee or hip surgery may influence different outcomes in different ways. To obtain a holistic picture of recovery, it would thus be better to look at a range of outcomes [11]. Given the study design of ALSWH, no objective measures on lower limb function were available. However, the PF subscale of the SF-36 asks about daily and mobility-related functioning, which patients and other stakeholders have identified as important outcomes for osteoarthritis management including joint replacement surgery [28, 29].

A final limitation is selective dropout. Participants who were still in the study at survey 6 (2011) were more likely to report hip or knee surgery during follow-up than participants who dropped out (online suppl. Table 1). This is likely explained by the shorter follow-up duration, resulting in less time to develop conditions requiring knee or hip surgery. Indeed, the main reason for dropout was death. Participants who remained in the study were more often physically active and less often needed help with daily tasks than those who dropped out (p < 0.001). However, sensitivity analyses including only those participants who completed all surveys showed similar results to the main analyses including participants who provided any data at any time point (online suppl. Table 2; Fig. 2). Hence, the influence of selective dropout on the findings is likely small. If selective dropout does play a role, it would have made the sample more homogenous and less likely to be able to differentiate between those with better and poorer resilience, thus leading to underestimation of the findings.

The finding that pre-surgery health and functioning were lower in women with poorer recovery trajectories after knee or hip surgery than in women with good recovery suggests that such information may be useful in developing pre-surgery prediction models for post-surgery outcomes. Such prediction models can then be used in triage for knee or hip surgery. However, to improve accuracy of such prediction models, we advise that such models are developed for specific patient groups and types of surgery.

The trajectory of PF differed between women with and without knee or hip surgery. A dip in PF was observed around the time of hip surgery. After surgery, PF remained lower than in women without surgery, but with a similar rate of decline from 3 years after surgery. In women with knee surgery, no dip in PF was observed and it remains unclear why this was the case. However, the level of functioning was lower at all times, with a similar rate of decline as in the women without surgery. When modelling long-term trajectories of PF in older surgery patients, the ageing-related decline in functioning should be taken into account. Women with high resilience to surgery whose PF recovered better than expected up to 3 years after surgery had fewer health problems and were more independent.

The research on which this paper is based was conducted as part of the Australian Longitudinal Study on Women’s Health by The University of Queensland and the University of Newcastle. We are grateful to the women who provided the survey data.

The study protocol was reviewed and approved by the University of Newcastle Human Research Ethics Committee (Approval No. H-076-0795) and The University of Queensland Human Research Ethics Committee (Approval No. 2004/HE000224). All participants provided signed informed consent.

The authors have no conflicts of interest to declare.

This work was supported by the Australian Government Department of Health and Aged Care.

All authors provided input on the study design, interpreted the results, and approved the final draft of the manuscript. G.P. and I.M. ran the analyses and drafted the manuscript. L.T. and R.M. provided critical feedback on the analysis plan and drafts of the manuscript.

ALSWH survey data are owned by the Australian Government Department of Health and Aged Care, and due to the personal nature of the data collected, release by ALSWH is subject to strict contractual and ethical restrictions. Ethical review of ALSWH is done by the Human Research Ethics Committees at The University of Queensland and the University of Newcastle. De-identified data are available to collaborating researchers where a formal request to make use of the material has been approved by the ALSWH Data Access Committee. The committee is receptive to requests for datasets required to replicate results. Information on applying for ALSWH data is available at https://alswh.org.au/for-data-users/applying-for-data/.