For this study, a subgroup of the prospective Huisartsen Onderzoek Netwerk Erasmus Universiteit Rotterdam (HONEUR) knee cohort was used; details on this cohort have been reported earlier (12). In brief, consecutive patients visiting their general practitioner with a new episode of knee symptoms were enrolled in the study and followed for 1 year. In this prospective cohort study, 40 general practitioners from 5 municipalities in the southwest region of The Netherlands participated, connected to the Erasmus Medical Center General Practitioner Research Network HONEUR and representing a total patient population of ∼84,000. Recruitment began in October 2001 in 1 municipality and a new municipality was added approximately every 3 months. All general practitioners continued to recruit until October 2003 (12).

New symptoms were defined as symptoms presented to the general practitioner for the first time. Recurrent symptoms for which the general practitioner was not consulted within the past 3 months were also considered to be new symptoms. Exclusion criteria were knee symptoms that required urgent medical attention (fractures, infection), patients with malignancies, neurologic disorders, or systemic musculoskeletal diseases (e.g., Parkinson's disease, rheumatoid arthritis, amyotrophic lateral sclerosis), as well as patients being incapable of understanding the ramifications of participation.

At baseline and at 1-year followup, patients underwent a standardized physical examination of their knee by trained physiotherapists. The physical examination at baseline was planned as close to the date of consultation of the general practitioner as possible. Disability and pain were assessed every 3 months by means of self-reported questionnaires.

For this study, all patients age ≥35 years with nontraumatic knee symptoms were included. At baseline and at followup, data on knee symptoms (duration, intensity), daily activities, and social circumstances were collected and a physical examination of the knee was performed.

Functional disability and pain were assessed both at baseline and at followup by self-report questionnaires containing the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) (13, 14), the Medical Outcomes Study Short Form 36 Health Survey (SF-36) (15, 16), the Knee Society Score (KSS) function questions (17, 18), the Lysholm Knee Scoring Scale (19–21), the Tampa Scale for Kinesophobia (assessed at baseline) (22, 23), and questions about experience of recovery or worsening (assessed at 1-year followup).

The physical examination assessed signs (e.g., swelling, temperature) and symptoms (e.g., function, pain) of the knee and hip. Further details about the physical signs and how they were elicited and scored are available from the corresponding author. For the outcome of persisting knee symptoms at 1-year followup, an additional question addressing experienced recovery or worsening, scored on a 7-point Likert scale, was added to the last questionnaire.

For the missing data, a multiple imputation strategy (multiple imputation by chained equations) was used (24). First, to assess which factors of the medical history and physical examination reported at baseline were associated with persisting knee symptoms (i.e., persisting or worsening knee symptoms) at 1-year followup, a univariate analysis was performed. The baseline factors that were analyzed were based on the literature (5, 10) and on experienced clinical relevance. The determinants were divided in 3 subgroups: patient characteristics, symptom characteristics, and physical examination findings.

Patients with persisting knee symptoms were defined as patients who, after 1 year of followup, experienced knee symptoms (somewhat better, no recovery, worse, much worse, or worse than ever) versus recovered patients who experienced knee symptoms (much better or no symptoms).

In the multivariate backward logistic regression analysis, factors with P < 0.2 in the univariate analysis were included (P entry 0.05, P removal 0.10). To assess the prognostic value of determinants with persisting knee symptoms, a prognostic model was built by backward logistic regression and the area under the receiver operating characteristic curve (AUC) was estimated. Three models were built: the patient characteristics model, the self-reported symptom characteristics model, and the physical examination findings model.

To assess the added predictive value of self-reported symptom characteristics, these factors were added to the model of patient characteristics; improvement was expressed as the difference between the AUCs. Adding the physical findings model to the model of patient and symptom characteristics assessed the added predictive value of determinants of physical examination. For this model, the AUC was also estimated. In addition, based on the age groups used in the American College of Rheumatology (ACR) clinical classification criteria of knee OA (25), we also performed these analyses separately for the patient age subgroups 35–49 years and ≥50 years.

We chose to dichotomize most variables because this allows estimation of odds ratio parameters through a logistic regression analysis (26), which are easier to interpret in clinical practice. However, the consequences of dichotomizing are an overall reduced statistical power, loss of information, and an increased probability of a Type II error (27, 28). SPSS software, version 11 (SPSS, Chicago, IL) was used to analyze the data.

A total of 549 patients were included. Their mean ± SD age was 53.8 ± 11.4 years, their mean ± SD body mass index (BMI) was 27.1 ± 4.3 kg/m², and 269 (49%) were women. Details on the baseline characteristics of the study group are presented in Table 1.

At 1-year followup, 480 (87.4%) persons were still available for the study; of these, 236 (49.2%) reported persisting knee symptoms. The 69 (12.6%) patients lost to followup showed no significant differences compared with those not lost to followup regarding baseline age, sex, BMI, KSS knee and function score, SF-36 score, comorbidity, and WOMAC scores. Of the persons lost to followup, reasons for no longer participating were lack of time/lack of interest (n = 36, 52.2%), severe comorbidity (n = 15, 21.7%), or treatment by an orthopedic surgeon (n = 4, 5.8%), and 14 (20.3%) patients provided no reason.

Multiple imputation was used to replace the missing values. There were 8 missing values for the dependent variable persistent knee symptoms. Of the patient characteristics, symptom characteristics, and characteristics of physical examination, the range of missing values was 3–20. An exception was the Heberden's nodes characteristic, for which there were 178 missing values. Further information about the proportion of missing data for each covariate is available from the corresponding author.

During 1-year followup, 43 (9.0%) patients underwent an operation for their knee. Of these, 18 (41.9%) reported persisting knee symptoms and 25 (58.1%) reported recovery (P = 0.26). Also during the 1-year followup, the WOMAC scores increased, with the largest increase at the 3-month followup measurement (Table 2).

The factors analyzed in the univariate analysis and their association with persisting knee symptoms are shown in Table 1. Of the patient characteristics, 7 variables were significant at P < 0.20: age >60 years, female sex, a low/moderate educational level, comorbidity of the skeletal system, kinesophobia (Tampa score >25), paid employment >8 hours per week, and sport participation.

Of the self-reported symptom characteristics, 12 variables were significant at P < 0.20: a warm knee, a swollen knee, crepitus of the knee, the presence of bilateral symptoms, duration of symptoms >3 months, feeling of giving way, limitation when walking stairs, recurrent symptoms, a history of nontraumatic knee symptoms, a history of traumatic knee symptoms, SF-36 physical component summary (PCS) score (continuous variable), and the SF-36 total score (continuous variable). The baseline WOMAC total score, WOMAC physical functioning score, WOMAC pain score, and WOMAC stiffness score were not associated with persisting knee symptoms.

Of the variables on physical examination, 7 variables were significant at P < 0.20: a swollen knee joint, crepitus with passive extension, crepitus with active extension, a bony swelling of the joint, pain of internal rotation of the hip, a restriction of internal rotation of the hip, and the presence of Heberden's nodes.

Of the patient characteristics, 4 variables remained in the multivariate prognostic model: age >60 years, a low/moderate educational level, presence of comorbidity of the skeletal system, and kinesophobia (AUC = 0.67, median predicted values 0.49, range 0.54, interquartile range [IQR] 0.28) (Table 3). Of the self-reported symptom characteristics, 4 variables remained in the model: the presence of bilateral symptoms, a history of knee symptoms (traumatic and nontraumatic), and duration of symptoms >3 months (AUC = 0.73, median predicted values 0.36, range 0.68, IQR 0.34). Of the variables on physical examination, only the crepitus of passive extension of the knee remained in the model (AUC 0.55, median predicted values 0.47, range 0.16, IQR 0.0).

To assess the added value of self-reported symptom characteristics on the model of patient characteristics, we added the model of self-reported symptom characteristics to the model of patient characteristics and calculated the AUC (Table 3). The strongest predictors in this new model were the symptom characteristics variables. The AUC improved, with an increase of 0.09 to 0.76 (median predicted values 0.42, range 0.83, IQR 0.37). We then added the variable of crepitus of passive extension of the knee to assess the added value of this variable, but the model did not improve any further (AUC 0.76, median predicted values 0.45, range 0.83, IQR 0.36).

For the age group 35–49 years, the factors analyzed in the univariate analysis and their association with persisting knee symptoms are shown in Table 1.

Of the patient characteristics, 3 variables were significant at P < 0.20: low/moderate educational level, comorbidity of the musculoskeletal system, and paid employment >8 hours per week. Of the self-reported symptom characteristics, 4 variables were significant at P < 0.20: duration of symptoms >3 months, a warm knee, a history of nontraumatic knee symptoms, and the presence of bilateral symptoms. Of the characteristics on physical examination, 3 variables were significant at P < 0.20: pain with active flexion of the knee, pain with active extension of the knee, and pain of internal rotation of the hip.

Of the patient characteristics, 2 variables remained in the multivariate model: low/moderate educational level and presence of comorbidity of the musculoskeletal system (AUC = 0.63, median predicted values 0.42, range 0.30, IQR 0.30) (Table 4). Of the self-reported symptom characteristics, 2 variables remained in the multivariate model: duration of symptoms >3 months and a warm knee (AUC = 0.64, median predicted values 0.31, range 0.99, IQR 0.67). Of the characteristics on physical examination, no variables remained in the multivariate model.

To assess the added value of self-reported symptom characteristics on the model of patient characteristics, we added the model of self-reported symptom characteristics to the model of patient characteristics and calculated the AUC (Table 4). The strongest predictors in this new model were a low/moderate educational level, duration of symptoms >3 months, and a warm knee. The AUC improved, with an increase of 0.08 to 0.71 (median predicted values 0.31, range 0.99, IQR 0.68).

In the age group ≥50 years, 7 patient characteristics variables were significant at P < 0.20: age >60 years, female sex, low/moderate educational level, comorbidity of the musculoskeletal system, kinesophobia, paid employment, and sport participation.

Of the self-reported symptom characteristics, 10 variables were significant at P < 0.20: duration of symptoms >3 months, a warm knee, crepitus of the knee, a history of nontraumatic knee symptoms, a history of traumatic knee symptoms, feeling of giving way, bilateral symptoms, SF-36 total score, SF-36 PCS score, and recurrent symptoms.

Of the characteristics of physical examination, 5 variables were significant at P < 0.20: crepitus of active flexion, pain when passive extension, a bony swelling of the joint, pain of internal rotation of the hip, and restriction of internal rotation of the hip.

Of the patient characteristics, 4 variables remained in the multivariate model: age >60 years, female sex, kinesophobia, and sport participation (AUC = 0.69, median predicted values 0.52, range 0.57, IQR 0.25) (Table 5). Of the self-reported symptom characteristics, 4 variables remained in the multivariate model: duration >3 months, a history of nontraumatic knee symptoms, recurrent symptoms, and bilateral symptoms (AUC = 0.76, median predicted values 0.42, range 0.85, IQR 0.64). Of the characteristics of physical examination, no variables remained in the multivariate model.

To assess the added value of symptom characteristics on the model of patient characteristics, we added the model of symptom characteristics to the model of patient characteristics and calculated the AUC (Table 5). In this new model, the strongest predictors were the variables of self-reported symptom characteristics. The AUC improved, with an increase of 0.11 to 0.80 (median predicted values 0.48, range 0.91, IQR 0.72).