Long-term clinical and economic outcomes of a short-term physical activity program in knee osteoarthritis patients

Abstract

Objective:

Physical activity (PA) in the US knee osteoarthritis (OA) population is low, despite well-established health benefits. PA program implementation is often stymied by sustainability concerns. We sought to establish parameters that would make a short-term (three-year efficacy) PA program a cost-effective component of long-term OA care.

Method:

Using a validated computer microsimulation (Osteoarthritis Policy Model), we examined the long-term clinical (e.g., comorbidities averted), quality of life (QoL), and economic impacts of a three-year PA program, based upon the SPARKS (Studying Physical Activity Rewards after Knee Surgery) Trial, for inactive knee OA patients. We determined the cost, efficacy, and impact of PA on QoL and medical costs that would make a PA program a cost-effective addition to OA care.

Results:

Among the 14 million with knee OA in the US, >4 million are inactive. Participation of 10% in the modeled PA program could save 200 cases of cardiovascular disease, 400 cases of diabetes, and 6,800 quality-adjusted life-years (QALYs). The program had an incremental cost-effectiveness ratio (ICER) of $16,100/QALY. Tripling PA program cost ($860/year) raised the ICER to $108,300/QALY; varying QoL benefits from PA yielded ICERs of $8,800/QALY-$99,900/QALY; varying background cost savings from PA did not qualitatively impact ICERs. Offering the PA program to any adults with knee OA (not only inactive) yielded $31,000/QALY.

Conclusion:

A PA program with three-year efficacy in the knee OA population carried favorable long-term clinical and economic benefits. These results offer justification for policymakers and payers considering a PA intervention incorporated into knee OA care.

Introduction

Physical activity (PA) is widely accepted as a key to promoting public health.(1–3) Despite the well-studied mental and physical health detriments associated with physical inactivity, increasing uptake of PA in the US has proven difficult, and levels of PA remain low.(4–6) Individuals with symptomatic knee osteoarthritis (OA), a painful and mobility-limiting chronic disease affecting about 14 million adults in the US,(7) may receive additional health benefits from PA, including reduced pain and improved physical function.(8, 9) However, low PA uptake in this population still leads to an estimated 7.5 million quality-adjusted life-years (QALYs) lost.(10)

Trials have evaluated PA-promoting interventions in the knee OA population, including interventions focused on health coaching and financial incentives, weight-loss paired with dietary components, education, and walking and conditioning exercise.(11–14) As of June 2019, there were over 100 recruiting and/or active trials registered in ClinicalTrials.gov evaluating PA in the US adult OA population.(15) Some exercise interventions in the knee OA population have been found to be cost-effective over a lifetime horizon, such as when in conjunction with dietary interventions or when applied to subsets of the knee OA population.(16–18) However, evidence for the long-term sustainability of PA levels achieved during short-term interventions is lacking.(17, 19–21) Consequently, stakeholders, policymakers, and payers may be averse to funding a PA program due to concerns of patients returning to inactivity soon after treatment completion.

In this model-based evaluation, we sought to assess the lifetime-horizon clinical and economic impacts arising solely from a short-term (three-year) PA intervention and to establish dimensions for key parameters-such as intervention cost and efficacy and impact of PA level on background medical costs and quality of life (QoL)-that would make even a PA program implemented for three years a cost-effective addition to management of knee OA in the long-term.

Methods

Analytic overview

We used a validated, widely-published state-transition computer microsimulation model, the Osteoarthritis Policy (OAPol) Model,(22–26) to evaluate the long-term clinical and economic impacts of a short-term PA intervention in conjunction with standard of care (SOC) for the US knee OA population. We simulated cohorts of inactive older adults with symptomatic knee OA and modeled them over their remaining lifetimes both with and without the PA intervention. To measure the long-term value of the PA intervention, we calculated incremental cost-effectiveness ratios (ICERs) as the difference in cost over the lifetime difference in quality-adjusted life years (QALYs) between the simulated scenarios with and without the PA intervention. We also evaluated how many QALYs and cases of CVD and DM would be saved if various proportions of the inactive OA population were to participate in the PA intervention.

We categorized PA by three levels-inactive (0–30 minutes of moderate-to-vigorous PA (MVPA)/week, insufficiently active (30–149 minutes/week), and active (150+ minutes/week)-guided by metrics for adults established in the Second Edition of the CDC Physical Activity Guidelines for Americans(2). To derive the prevalence of each PA level in the OA population, we used accelerometry data from the Osteoarthritis Initiative (OAI), a longitudinal study of adults with symptomatic knee OA.(27) We defined efficacy of the PA intervention as its likelihood of increasing PA by at least one level. We modeled the base-case PA intervention after the program in the Studying Physical Activity Rewards after Knee Surgery (SPARKS) trial,(11, 17) which included personalized telephonic coaching, financial incentives, and a personal activity monitor and defined PA by both number of steps/day and bouts of steps/minute consistent with three metabolic equivalents of task.

As recommended by the 2^nd Panel on Cost-Effectiveness in Medicine, we discounted QALY and cost outcomes at 3% per year(28, 29), and we reported cost outcomes in 2017 USD. We varied a spectrum of PA intervention specifications and PA-related background parameters to establish the ranges of their values within which a PA program would be a valuable, cost-effective addition to knee OA care. We defined “valuable” as having ICERs lower than several established willingness-to-pay (WTP) thresholds: $100,000/QALY, frequently recommended and used in economic evaluations in the US,(29–31) and $62,600 and $187,800/QALY, which correspond to 1x and 3x the 2018 US per capita GDP,(32) as recommended by the World Health Organization.(33)

The OAPol Model

The OAPol Model is a validated Monte Carlo simulation of the natural history and treatment of knee OA.(16, 22, 23) Subjects in the model each move through health states on an annual basis according to a set of transition probabilities. Health states in the model are defined by PA level, presence of comorbidities (cardiovascular disease (CVD) and type 2 diabetes mellitus (DM)), pain level, Kellgren-Lawrence (KL) grade, and obesity level. Transition probabilities were derived from secondary analysis of national cohorts or published literature and have been described elsewhere.(7, 16, 23)

Each year, subjects accrue QoL utility, a value ranging from 0 to 1 (1 represents a year experiencing perfect health and 0 a year in death), and medical costs due to OA treatment and background healthcare. Both annual QoL utility and background medical costs are impacted by PA level in the model.

Interventions in the OAPol Model are defined by efficacy, duration, and cost. Efficacy of the PA intervention was defined as the likelihood of increasing PA level, which in turn leads to improved QoL and lower background medical costs in the model. OA-focused interventions carry the likelihood of decreasing pain and, as a result, improving QoL. Interventions in OAPol may include risk of toxicities, which themselves lead to QoL decrements and cost increases. The SOC treatment for symptomatic OA in the OAPol Model includes pharmacologic pain management, physical therapy, and the eventual option of TKR. The details of this SOC treatment course have been described in previous publications and are summarized in Figure 1.(22–26) Section 1 of the Technical Appendix (TA) contains further detail of the OAPol Model structure and assumptions.

Figure 1. Flowchart of Model Interventions.

After model initialization, base case (“PA Program”) subjects simultaneously were offered the PA intervention in the model and the standard of care (SOC) treatment path (including non-surgical pain management and offer of TKR). This is indicated by the “PA Program” brackets on the right-hand side of the flow diagram, which encompasses both paths in parallel. Subjects remained on the PA program for three years in parallel with the SOC sequence. The “SOC” brackets indicate that the SOC comparator cohort to the “PA Program” cohort was only offered the SOC path. The SOC path began with a non-surgical treatment regimen for osteoarthritis pain; this treatment included NSAIDs, acetaminophen, office visits, and physical therapy. Following the non-surgical treatment, subjects were offered a TKR if they had a KL OA grade of at least 3 and a WOMAC pain level greater than 40. Subjects left the non-surgical pain management treatment if they did not achieve sufficient pain reduction or maintain reduced pain. Those in the SOC-only comparator group entered the non-surgical pain management treatment upon model initiation. All subjects left the modeling sequence upon death, which was determined probabilistically.

The OAPol Model aggregates the results from large numbers of individual simulations to develop stable population estimates of outcomes. Output includes average discounted QALE and discounted costs, as well as the cumulative incidence of CVD and DM, for all subjects over their remaining lifetime following model run initiation.

Input parameters: Cohort characteristics and natural history

Initial cohort characteristics for the current analysis were based on the demographic and clinical characteristics of persons with knee OA, with a mean (SD) age of 55 (5), mean symptomatic pain of 30 (per the Western Ontario and McMaster Universities Osteoarthritis Index(34) (WOMAC); range of 0–100, with 100 as worst pain), and KL grade of 2. We assumed that all subjects were inactive at baseline.

Baseline distributions of sex and obesity level and prevalence of all comorbidities (CVD, DM, cancer, non-OA musculoskeletal disease, and chronic obstructive pulmonary disease) were derived from 2014–16 National Health and Nutrition Examination Survey (NHANES) and CDC Lifetables(35, 36); comorbidity prevalence was stratified by age, sex, and obesity and was adjusted for lack of PA (see TA, Section 2, for derivation details). We derived initial pain distribution from baseline data from the OAI.(37) QoL utility was stratified by age, BMI, number of comorbidities, and level of pain, and it was also derived from the OAI.(27) Background medical costs were stratified by age, obesity level, and number of comorbidities and were derived from the NHANES, the Medicare Current Beneficiary Survey, and Pope et al.(36, 38, 39) We also adjusted background costs for lack of PA, described in Section 3 of the TA.

We report PA-related input values and natural history parameters in Table 1. We derived annual background probability of decreasing PA level from the OAI accelerometry data and did not include an annual probability of increasing PA level in the absence of the PA intervention, as this value was near 0 in the OAI dataset.(27)

Table 1.

Model Inputs Relevant to PA. Table 1 displays key model inputs relating to PA, including the mechanisms by which PA level influences subjects’ background parameters and the parameters of the PA program itself.

PA-based increment to QoL (95% CI)(27)
Inactive	0
Insufficiently Active	0.019 (0, 0.04)
Active	0.037 (0.014, 0.060)

PA-based annual cost savings(40, 41)
Inactive	$0
Insufficiently Active	$313
Active	$714

Relative risk of comorbidity incidence
	Cancer	Diabetes	Cardiovascular disease
Inactive	1	1	1
Insufficiently Active	0.773	0.634	0.709
Active	0.758	0.520	0.583

Annual background probability of decreasing PA level(27)
Insufficiently active to inactive	0.0831
Active to insufficiently	active 0.0801

Probability of failure* on PA intervention(11)
Starting PA level	First year	Subsequent years
Inactive	0.62	—
Insufficiently active	0.57	0.06
Active	0.50	0.21

^*First year failure is defined as either not advancing by one PA level, if starting at inactive or insufficiently active, or not remaining active the following year if already active. Therefore, no subjects who remain inactive and advance to be evaluated for subsequent year efficacy. Subsequent year failure is defined only as decreasing PA level.

PA level impacted risk of developing CVD and DM based upon relative risk values derived from a review we conducted of prospective studies of cardiorespiratory fitness level and development of disease. We derived QoL utility increments corresponding to active and insufficiently active levels of PA using the OAI accelerometry data.(27) Finally, we assumed, based on the published literature, that higher PA level was associated with lower background medical costs, and we associated cost savings with being insufficiently active or active.(40, 41)

Input parameters: PA intervention

The PA intervention was offered over the course of three years with the purpose of increasing or maintaining high PA levels. We offered the PA program alongside SOC for knee OA.

We defined success in the first year of the PA intervention as increasing a subject’s PA level; subjects could fail in years two and three by decreasing PA level. We derived probabilities of intervention success and failure from the SPARKS trial data, reported in Table 1.(11) Further details of PA intervention efficacy derivations are presented in the TA (Section 4). The annual PA program cost was $287, which represented the average costs (health coaching, wearable commercial accelerometer, and financial incentives) from the SPARKS study of an exercise intervention for the knee OA population undergoing eventual TKR.(11, 17)

We made several assumptions regarding the structure of the PA program. The program did not lead to any major or minor adverse events (as it was based upon walking), and it did not have a direct impact on subjects’ pain levels (to remain conservative). All subjects accrued program costs for three years, regardless of whether they were able to improve PA levels, to reflect a realistic scenario for a payer in which participants do not formally withdraw from a program despite non-optimal adherence.

To maintain the conservative nature of this analysis, we assumed that all subjects reverted to their pre-treatment PA levels following the conclusion of the three-year PA intervention period. As a result, subjects returned to the risks of CVD and DM, along with the annual QoL utility and background medical costs, associated with those pre-treatment PA levels. This allowed us to examine the long-term benefits of a short-term PA program.

Population benefits

We estimated the number of QALYs that would be saved and cases of CVD and DM that would be averted if 5%, 10%, and 15% of the inactive knee OA population were to participate in the PA intervention. To derive the size of the population of US adults who were inactive with knee OA, we used OAI accelerometry data (27) in conjunction with population size data stratified by sociodemographic characteristics from CDC Wonder(42) and OA prevalence data from Deshpande et al.(7) Using these data, we calculated the total number of potential QALYs saved and CVD and DM cases averted with the PA intervention in the US OA population by multiplying per-person estimates obtained from the OAPol model by the corresponding true population size. The details of these calculations are outlined in Section 5 of the TA.

PA and PA program characteristics examined

We varied the following parameters by a wide range of possible values (presented below) in deterministic sensitivity analyses: (1) annual background cost savings attributable to high PA level, (2) annual cost of PA program, (3) annual background QoL increments attributable to high PA level, and (4) level of activity in the population eligible for the PA program. The base case values for these parameters are reported in Table 1.

We decreased the annual background medical cost savings attributable to high PA level by 50% and by 100% (details presented in TA Section 3). We varied the annual cost of the PA programs to be 2–5 times greater than the base case. We varied the annual increments to QoL utility due to PA level according to their upper and lower 95% confidence interval (CI) bounds, reported in Table 1. Finally, we examined how offering the PA intervention to any US adults with symptomatic knee OA, regardless of baseline PA level, would affect the program’s value, defined by its ICER. We report distribution used of PA level in the US OA population in the TA, Section 6.

We conducted several sets of two-way sensitivity analyses examining how simultaneously varying pairs of key parameters – (1) annual background cost savings due to high PA, (2) cost of the PA program, and (3) annual QoL utility increment due to high PA – would impact the cost-effectiveness of the PA program.

We also conducted probabilistic sensitivity analyses, varying the following four parameters: (1) annual background cost savings due to high PA, (2) cost of the PA program, (3) annual QoL increment due to high PA, and (4) efficacy of the PA program. Annual cost savings due to PA and cost of the PA intervention were varied using a log normal distribution, annual QoL increment due to PA was varied using a normal distribution, and efficacy of the PA intervention was varied using a beta distribution. We report further details of these distributions and analyses in Section 7 of the TA. We used a cost-effectiveness acceptability curve to visualize the proportion of ICERs from all probabilistic scenarios (out of 500) that fell under a range of WTP thresholds.

Results

Base Case

Adding the PA intervention to SOC for knee OA yielded a difference in QALY of 1.7 QALYs per 100 participants and a difference in cost of $274, resulting in an ICER of $16,100/QALY (Table 2).

Table 2.

Base case and one-way deterministic sensitivity analyses Table 2 reports the average discounted life expectancy, discounted cost, and incremental cost-effectiveness ratio (ICER) for the PA program with SOC compared to SOC alone for the base case analysis and all one-way deterministic sensitivity analyses.

Scenario	Life Expectancy	Cost	Δ Cost*	Δ QALY*	ICER ($/QALY)
Base Case
SOC alone (Ref**)	12.73	160,413
PA program + SOC	12.75	160,688	274.42	0.017	16,142
Sensitivity Analyses
Cost of PA program
2x cost	12.75	161,509	1095.08	0.017	64,416
3x cost	12.75	162,364	1950.22	0.018	108,346
4x cost	12.75	163,170	2756.57	0.017	162,151
5x cost	12.75	164,004	3590.68	0.017	211,216
PA-related background medical cost savings
Ref	12.73	157,937
50% reduced+	12.75	158,359	422.03	0.017	24,825
Ref	12.73	155,501
100% reduced	12.75	156,027	526.55	0.017	30,974
PA-related QoL increment
Ref	12.73	160,417
Upper bound QoL++	12.77	160,700	282.33	0.032	8,823
Ref	12.74	160,414
Lower bound QoL++	12.74	160,713	299.61	0.003	99,870
Eligibility for PA program
Ref	13.08	151,133
Full OA population, all PA levels	13.09	151,536	402.54	0.013	30,965

^*ΔCost and ΔQALY are relative to SOC treatment regimen with the same natural history characteristics.

^**For each pair of runs, the “Ref” comparator mirrors the cohort parameters varied but is run with only SOC.

⁺Reduced savings refers to the decrements to annual background medical costs associated with being active and insufficiently active. Under 100% reduced savings, background annual medical costs were the same for all subjects, regardless of PA level.

⁺⁺Lower and upper QoL bounds refer to the lower and upper 95% confidence interval values for the increments to applied annual QoL utility increment associated with being active and insufficiently active.

We estimated that 4.1 million US adults aged 45+ with asymptomatic knee OA were inactive, defined as <30 minutes of MVPA/week. If 5% of this inactive knee OA population took part in such a PA program, 105 cases of CVD and 188 cases of DM could potentially be averted, and 3,400 potential QALYs could be saved. If 10% of that population participated in the PA program with SOC, 210 cases of CVD, 377 cases of DM, and 6,800 QALYs could be saved. If 15% of that population participated in the PA program with SOC, 314 cases of CVD, 603 cases of DM, and 10,200 QALYs could be saved.

One-Way Deterministic Sensitivity Analyses

Results of the one-way deterministic sensitivity analyses are reported in Table 2 and visualized in Figure 2. All one-way sensitivity analyses, except for the analyses varying cost of the PA intervention, yielded ICERs less than the $100,000/QALY WTP threshold. Reducing the background PA cost savings by 50% and 100% resulted in ICERs of $24,800/QALY and $31,000/QALY, respectively. Increasing the annual cost of the PA intervention 2- to 5-fold yielded ICERs that ranged from $64,400/QALY to $211,200/QALY, surpassing the WTP threshold of $187,800/QALY (3x the 2018 GDP) only if the cost of PA intervention was increased to 5 times the base case value. Setting the QoL increment due to PA to its upper 95% CI bound yielded an ICER below the more conservative WTP threshold of $62,600/QALY, while setting the QoL increment equal to its lower 95% CI bound led to an ICER of $99,900/QALY. When we considered offering the PA intervention to any knee OA patient, regardless of initial PA level, the resulting ICER was $31,000/QALY.

Figure 2. Tornado diagram of one-way sensitivity analyses.

Figure 2 depicts the resulting ICERs from the one-way deterministic sensitivity analyses conducted. The dashed line represents the base case ICER of $16,142/QALY. The dots in each bar represent each variation of the parameter represented by the bar, along with that parameter’s base case value (aligned with the dashed vertical line). From left to right, we varied the annual QoL increment due to PA level to equal its upper 95% CI and its lower 95% CI (values ranged from 0 to 0.040 for insufficiently active and from 0.014 to 0.060 for active); we varied the annual cost savings due to PA level to be reduced by 50% and by 100% (values ranged from $0 to $313 for insufficiently active and from $0 to $714 for active); we varied the cost of the PA intervention to be 2x, 3x, 4x, and 5x the base case value (values ranged from $287 to $1,435).

Two-Way Deterministic Sensitivity Analyses

We depict the results of the two-way sensitivity analyses as a heat map in Figure 3. When simultaneously varying annual background medical cost savings due to PA level (greater cost savings with higher PA) and cost of the PA program, ICERs ranged from $16,100 (both parameters at base case value) to $226,700/QALY (annual cost savings due to PA level decreased by 100% and cost of PA program increased 5-fold). Scenarios with the PA program cost increased at least three-fold and the annual medical cost savings due to PA level reduced by any amount had ICERs over the $100,000/QALY WTP threshold; increasing PA program cost four-fold exceeded the $187,800/QALY WTP threshold for all scenarios that decreased annual cost savings due to PA except from its base value.

Figure 3. Two-way sensitivity analysis heat maps.

This heat map depicts the results of the two-way sensitivity analyses varying cost of the PA program, annual QoL increment due to high PA level, and annual background medical cost savings due to high PA level. ICER units are US dollars/quality-adjusted life-year. *$62,606 is equal to the 2018 US per capita gross domestic product (GDP). ^†$187,817 is equal to 3x the GDP.(32, 33)

When simultaneously varying annual background medical cost savings due to PA level and annual QoL increment due to PA level (greater QoL increment with higher PA), ICERs ranged from $8,800 (annual cost savings due to PA at base case value and QoL increment due to PA equal to upper 95% CI) to $146,700/QALY (annual cost savings due to PA 100% reduced and QoL increment due to PA set to lower 95% CI). These ICERs only surpassed the $100,000/QALY threshold when using the lower 95% CI for annual QoL increment due to PA and decreasing annual cost savings due to PA below base value. ICERs never surpassed the $187,800/QALY WTP threshold.

When simultaneously varying cost of the PA program and annual QoL increment due to PA level, ICERs ranged from $8,800 (cost of the PA program at base case value and QoL increment due to PA set to upper 95% CI) to $905,400/QALY (cost of the PA program increased 5-fold and QoL increment due to PA set to lower 95% CI). Scenarios with the PA program cost doubled fell below the $100,000/QALY WTP threshold when annual QoL increment due to PA equaled the base case value or its upper 95% CI. All PA program cost scenarios fell below the $187,800/QALY WTP threshold when paired with the upper 95% CI scenario for annual QoL increment due to PA level.

Probabilistic Sensitivity Analyses

Results of PSA are depicted in the cost-effectiveness acceptability curve (Figure 4). The PA program with SOC as compared to SOC alone was cost-effective under a WTP threshold of $100,000/QALY in 87% of all scenarios. The PA program was cost-effective under a WTP threshold of $62,600/QALY in 79% of all probabilistic scenarios and was cost-effective under $187,800/QALY in 94% of scenarios. Twenty-four percent of the 500 total scenarios were cost-saving (incorporating the PA program both decreased average costs and increased average QALYs as compared to SOC alone), and 1% were dominated (incorporating the PA program both increased average costs and decreased average QALYs as compared to SOC alone).

Figure 4. Cost-Effectiveness Acceptability Curve with Probabilistic Parameter Variation.

The cost-effectiveness acceptability curve in Figure 4 graphically depicts the probability (y-axis) that the ICER for the PA intervention plus SOC as compared to SOC alone falls under various WTP thresholds, which increase along the x-axis. These curves represent the ICERs resulting from 500 model runs probabilistically simultaneously varying the following parameters: (1) annual background cost savings due to PA, (2) cost of the PA intervention, (3) annual QoL increment by PA level, and (4) efficacy of the PA intervention. The dashed vertical lines on the graph represent the three WTP thresholds we evaluated in our analyses: $62,600/QALY (1x the 2018 US per capita GDP), $100,00/QALY, and $187,800/QALY (3x the 2018 US per capita GDP).

Discussion

We found that a three-year PA intervention, for which the benefits were only realized during the three-year program period, was a valuable addition to knee OA care, with an ICER of $16,100/QALY when the program was offered only to the inactive knee OA population. When eligibility for such a PA program was extended to anyone with knee OA, the ICER increased slightly to $31,000/QALY.

These results add further support to the cost-effectiveness literature justifying the incorporation of a PA program into knee OA care. The intervention examined in the IDEA Trial, a study of a diet and exercise program in obese and overweight adults with knee OA, was found to be cost-effective, with a base case ICER of $34,100/QALY from the healthcare perspective.(16) The simulation of IDEA included annual costs for personnel ($328 in year 1, $281 in subsequent years), meal replacements ($455 once), and gym memberships ($600/year); this sums to a higher intervention cost than our $287/year, as we modeled a remote intervention focused only on PA via health coaching, financial incentives, and wearable accelerometers.(12)

A cost-effectiveness analysis of the financial incentives and health coaching program in the SPARKS trial, from which we derived cost and efficacy of the PA program in the current analysis, found the intervention to have an ICER of $57,200/QALY when offered, with no limit on duration of efficacy, to a population of individuals post-TKR.(17)

Analyses in the general population have evaluated alternative types of PA interventions and found some to be cost-effective and some not.(43, 44) A review of interventions based in the community and structured around social support found ICERs ranging from $14,300-$68,600/QALY, while interventions focused on individually-adapted health behavior ranged from $29,800-$46,900/QALY.(44) Another review of PA-promoting programs found that cost-effectiveness varied widely depending on program design, from $79,000(AUD; $51,400 USD))/disability-adjusted life-year (DALY; representative of one lost year of life in a healthy state) for physician referral to an exercise physiologist, to $3,000(AUD; $2,000 USD)/DALY for an internet-based intervention, to pedometer- and media-based interventions having a high likelihood of being cost-saving.(43, 45) Thomas et al. evaluated a two-year home exercise intervention aimed at reducing knee pain and concluded that the program would be cost effective with a willingness to pay £8,000 ($12,500 USD) for at least a 50% improvement in knee pain.(46) Given this variation, it could be useful to design future analyses evaluating these and other styles of PA intervention in the knee OA population.

The results of the current analysis should be interpreted in the context of some limitations. We derived the base case parameters for the PA program from a trial (SPARKS) focused on TKR recipients and applied these parameters to the broader population of inactive individuals with knee OA. Given this, we conducted sensitivity analyses to examine the robustness of our results. We probabilistically varied PA intervention efficacy to account for the fact that patients who undergo TKR may experience greater or smaller PA increases post-procedure from their higher-pain/lower-PA baseline levels than those of individuals who do not undergo TKR. The results of the probabilistic sensitivity analysis confirmed the robustness of our base case results. We also deterministically varied annual background cost savings and QoL increments attributable to high PA level to capture the potential differential in ongoing healthcare costs and QoL for individuals undergoing TKR and those not. The SPARKS trial also evaluated a remote PA promotional program, and remote programs are typically less expensive types of interventions.(12, 44) However, a remote program may be most readily applicable and easily disseminated to an inactive, older knee OA population.

In addition, we used multiple data sources to derive PA parameters used to describe the OA population. Deriving PA inputs from OAI, a large multicenter knee OA cohort,(27) may limit generalizability of results, as the OAI is not necessarily a representative sample of the entire US knee OA population. We also made the conservative assumption that PA level did not have a direct impact on OA pain.

Even under the most conservative assumption that participants revert completely to their baseline PA levels by the third year following initiation of a PA intervention, our results show that such a PA intervention would still be cost-effective over a lifetime horizon. The QoL benefits, background cost savings, and shifts in risk of chronic disease and healthcare utilization associated with the PA intervention period are significant enough to make an impact compared to SOC treatment alone.

Our findings offer additional insights into the value of PA programs and provide evidence that even a short-term PA program with a limited duration of benefits could be cost-effective in the knee OA population.