Development and feasibility of a personalized, interactive risk calculator for knee osteoarthritis

We used data generated by the Osteoarthritis Policy (OAPol) Model to create a computer-based risk calculator for knee OA (OA Risk C). OA Risk C allows users to enter their demographic and risk factor information, including family history of osteoarthritis, exposure to occupational risk factors (such as squatting and kneeling), and history of knee injury. Users are matched to one of 2016 OAPol simulations, which provides an estimate of their 5, 10, 15, 20, 25, 30-year, and lifetime risk of knee OA and total knee replacement (TKR).

OAPol is a validated, published, state-transition, Monte Carlo model of the natural history and management of knee OA [5, 6, 26, 27]. “State transition” implies that the model follows each person’s history as a sequence of annual transitions from one health state to another. Health states are designed to be predictive of clinical prognosis related to knee OA, risk for developing comorbidities, and mortality. The OAPol Model considers five major comorbidities (cardiovascular disease, diabetes mellitus, cancer, chronic obstructive pulmonary disease, and other musculoskeletal diseases). The prevalence and incidence of each of these comorbidities are stratified by age, sex, race/ethnicity, and obesity. Prevalence of cardiovascular disease is further stratified by diabetes mellitus status. Persons who develop comorbidities have higher risk for death. The explicit accounting for competing risks from other comorbidities is a special feature of the OAPol model.

Each person is followed from the time of entry into the OAPol Model until death. Upon entry into the model, a person is randomly assigned age, sex, race/ethnicity, and body mass index (BMI) from a set of user-specified probability distributions. At the start of each annual cycle, the model records the patient’s age, obesity status (defined by BMI), symptomatic knee osteoarthritis status, and presence of specific comorbidities. The model then uses these characteristics, in addition to sex and race/ethnicity, to determine the probabilities that indicate development of symptomatic knee osteoarthritis, progression of symptomatic knee osteoarthritis, development of a new comorbidity, and death in the subsequent year. Transition probabilities are derived from literature or secondary analyses of data from national- or population-based cohorts and translated into risk functions for the model. Details of transitional probabilities derivation have been previously published [26]. Upon the person’s death, summary statistics are recorded and a new person enters the model. A large number of individual simulations are aggregated to obtain stable estimates of OA and TKR rates. The size of the hypothetical cohort, N, is selected to ensure stable population estimates. For the purposes of establishing stable estimates for risk of knee OA, we used N = 1,000,000 for each model run.

We conducted 2016 model simulations using each unique combination of seven input parameters. These included starting age (25–45), sex (male, female), race/ethnicity (White, Black, Hispanic), obesity status (obese, non-obese), family history of knee OA (present, absent), occupational exposure to OA risk (present, absent), and history of knee injury (present, absent) (Table 1).

We defined incident knee osteoarthritis as Kellgren-Lawrence (K-L) grade 2 accompanied by knee pain. Base incidence rates were derived from the National Health Interview Survey and stratified by age, sex, and obesity. Details on incidence rates derivation have been previously published [28]. Upon development of symptomatic knee OA, subjects could progress through K-L grades 2 through 4 based on probabilities derived from the Johnston County Osteoarthritis Project, a cohort study of OA [29] and calibrated to published rates of annual knee OA progression [30]. Progression probabilities were stratified by sex, obesity, and current K-L grade (Table 1). Annual incidence of TKR was derived from two multi-centered longitudinal studies: MOST (Multicenter Osteoarthritis Study) and OAI (Osteoarthritis Initiative) [31, 32]. Further details of incidence rate derivation have been published elsewhere [7].

To account for increased risk of developing knee OA for those with a family history of knee OA, occupational exposure, obesity, or history of knee injury, we used the published data from Zhang et al. [33]. Based on the data from Zhang et al., knee injury increased OA incidence by a factor of 2.39. Family history of OA increased OA incidence by a factor of 1.72, and occupational risks by a factor of 1.28 (Table 1). Risk factors were assumed to be independent and multiplicative. As noted above, age, sex and obesity status were estimated directly based upon National Health Interview Survey data [28]. For example, to incorporate the impact of obesity, we first derived the incidence of knee OA among those who are not obese. This was done by dividing the age/sex stratified incidence of knee OA by the sum of the prevalence of obesity multiplied by the increased risk of OA due to obesity and prevalence of non-obesity. To ensure that we did not overinflate the risk of OA among obese persons, we applied increase in risk to the incidence of OA among those who are non-obese persons, derived as described above. We used the incidence rate among non-obese individuals as the basis for the risk adjustment for other risk factors.

OA Risk C users input demographic and risk factor information, including age, sex, race/ethnicity, height, and weight. Questions about risk factors are asked in such a way that someone unfamiliar with OA could answer them. Family history of osteoarthritis is determined by asking: “Do your parents, siblings, or grandparents have any of the following: Arthritis (osteoarthritis), Knee or hip replacement, Finger nodes?” Users are asked to check all that apply, including “Don’t Know” (Fig. 1). Users have the option of viewing a popup image of finger nodes if they are unfamiliar with the term. If the user selects any of the first three options (arthritis, knee or hip replacement, finger nodes), he or she is considered to have a family history of OA. Occupational risk is determined using the question: “Have you been exposed to occupational risk factors associated with knee OA such as kneeling, squatting, or lifting?” History of knee injury is determined by asking: “Have you previously had a serious knee injury that limited your ability to walk for at least 7 days?” This question conforms to the definition of knee injury used in the Osteoarthritis Initiative, a multi-center cohort study of knee OA [33].

Users are shown the results of their risk calculation in several different ways. The first page contains a simple graphical representation of their 5, 10, 15, 20, 25, 30-year, and lifetime risk of developing knee osteoarthritis and total knee replacement (Fig. 2). We also present risk information in icon array form, which users can access by clicking “What does my risk mean?” An example of this pictorial representation of the risk of developing knee OA is shown in Fig. 3. Then, users are taken to the interactive component of the risk calculator, where they can learn how different factors contribute to their risk of developing knee OA. Users are able to adjust their risk factors on an interactive graph. Four dropdown menus allow users to add or remove each risk factor and see how their risk assessment changes (Fig. 4). For example, a non-obese person can view their risk of developing osteoarthritis if they were obese. Users are also able to see which component of their risk is due to added risk factors, and which component simply reflects the underlying risk experienced by someone of their demographic group with no additional risk factors. As on the previous page, users are given the option to view this information in icon array form. Finally, users are shown a graphical comparison of how their personal risk of developing knee OA and undergoing TKR compares to that of the average American with no risk factors (Fig. 5).

In order to determine the feasibility and acceptability of the OA Risk Calculator, we enrolled 45 subjects at a primary care clinic within a tertiary medical center. Both patients and non-patients in the waiting room of the clinic were invited to participate. Those who spoke English, were between the ages of 25 and 45 years, had not been diagnosed with arthritis by a physician, had access to a computer and the Internet, and were willing to provide their email address were eligible for the study. After agreeing to participate, eligible subjects read and signed a digital consent form. They were given a convertible laptop (which functions as both a laptop computer and tablet) and instructed to follow the directions on the computerized risk calculator until the completion page appeared. Because the aim of pilot testing was to determine the usability of the calculator, the research assistant administering the calculator asked subjects to refrain from commenting or asking questions until the end of the survey if possible. After using OA Risk C, we provided study participants with a brochure about OA that included suggestions for minimizing exposure to modifiable OA risk factors. Patients were given a 2-h parking voucher for their participation or an Amazon gift card of equivalent value ($9) if they did not park.

To measure feasibility and acceptability of the calculator, we asked subjects to rate its ease of use and comprehensibility on a 5-point agreement scale with the following options: Strongly Disagree, Disagree, Neutral, Agree, and Strongly Agree. Participants were also asked for suggestions for improvement and general feedback. This study was approved by the Partners Healthcare Institutional Review Board (protocol 2014P000844).