Knee pain dilemma and the initial step to predicting diagnoses in general practice: a cross-sectional study

Introduction

Knee pain is a common reason for patients to consult a general practitioner (GP), but arriving at an accurate diagnosis poses a challenge for GPs.^1–3 However, a correct diagnosis is crucial as it not only guides appropriate management but also informs the prognosis. The frequency of knee pain referrals from general practice to secondary care is high.⁴ As knee pain is associated with age, the aging population puts GPs at the forefront of addressing the complexities associated with knee pain.⁵ Among these, the multifaceted nature of knee complaints, ranging from acute injuries to chronic conditions, the diversity and subjectivity of symptoms and signs, overlapping clinical presentations, and different diagnostic sets of criteria in international clinical guidelines contribute to the difficult diagnostic landscape faced by GPs.^5,6 The latter becomes evident when examining three frequently employed international guidelines for knee osteoarthritis, where each guideline employs a distinct age threshold.⁵

To support GPs with this diagnostic dilemma in patients with knee pain, a prediction model could be a useful diagnostic decision support tool, however, this has not yet been developed. This study marks a first step to simplify the diagnostic dilemma in knee pain with the aim of illuminating the path to a more predictive diagnostic approach that supports the GP in selecting the most appropriate management and informs the prognosis. Therefore, the objective of this study is to explore the most common established diagnoses in patients with knee pain referred by their GP to a primary care orthopaedic clinic, and investigate the association of these diagnoses with age, given that age is a recognised risk factor for various knee conditions and commonly used in guidelines by presenting age thresholds. Besides, the age of each patient is ascertainable without the need for history taking and a physical examination.

Methods

We conducted a cross-sectional study on medical records of patients with unilateral knee pain in the Netherlands, referred by their GP to a primary care orthopaedic clinic between January 2021 and June 2023, involving all patients in that period for whom the GP faced a diagnostic problem. This clinic provides care for low-complexity cases for which the referring GP does not expect a surgical indication. It offers personalised treatment recommendations based on diagnosis, often advising physiotherapy while also providing lifestyle guidance, injections (in-clinic or by their own GP), or specialist referrals if needed. GPs refer patients with non-traumatic knee complaints when surgery is unlikely and no magnetic resonance imaging (MRI) is required. All patients were seen by a dedicated team consisting of a GP with a special interest in musculoskeletal disorders, a dedicated physiotherapist, and an orthopaedic surgeon. Point-of-care ultrasound was used for initial imaging, following standardised criteria.^7,8 The orthopaedic surgeon established diagnoses based on patient history, physical exams, and imaging results, following current guidelines when applicable.

Descriptive statistics for numerical variables involved mean and standard deviation (s.d.), while for categorical variables frequency counts and percentages were used.

Age was categorised into 5-year thresholds for the most prevalent diagnoses, and we calculated sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and numbers of true negatives, true positives, false negatives, and false positives for each threshold. Univariate odds ratios (ORs) with 95% confidence intervals (95% CI) were determined using logistic regression analysis. Receiver operator curve (ROC) analysis was used to determine the optimal threshold value of patients’ age predictive for the diagnosis, corresponding to the value maximising the Youden Index. Bootstrapping with 2000 stratified replications was used to determine 95% CIs. The area under the ROC (AUC), including 95% CI (DeLong test), was calculated as an overall measure of discriminative ability. Data were analysed using SPSS (IBM Corp, Armonk, NY, USA, version 27.0) and R (R Core Team, Vienna, Austria, version 4.3.1)

This study was approved by the Medical Ethics Committee of Maastricht University (FHML-REC/2023/043).

Results

We included 627 patients with a mean age of 53.0 years (s.d. 16.9) of whom 56% (352/627) were female. Sixty-nine percent of the patients (430/627) were diagnosed with osteoarthritis of the knee, 9% were diagnosed with patellofemoral pain (54/627), followed by a meniscal lesion in 5% (29/627). All other diagnoses, e.g. tendinopathies or ligament injuries, had frequencies lower than or equal to 3%.

Table 1 displays the sensitivity, specificity, PPV, NPV, and ORs per 5-year age threshold for knee osteoarthritis, patellofemoral pain, and meniscal lesion. The ROC with AUC, optimal age threshold, and sensitivity and specificity corresponding to the determined age threshold for these conditions is depicted in Fig. 1. The optimal age threshold for osteoarthritis of the knee was 49.5 years (95% CI: 44.5–51.5) with an AUC of 0.90 (95% CI: 0.87–0.93), for patellofemoral pain 47.5 years (95% CI: 39.5–51.5) with an AUC of 0.87 (95% CI: 0.83–0.91), and for meniscal lesions 52.5 years (95% CI: 35.5–60.5) and an AUC of 0.72 (95% CI: 0.64–0.81).

Fig. 1.

Receiver operating characteristic curve for knee osteoarthritis, patellofemoral pain, and meniscal lesion, showing the optimal age threshold per disorder. AUC, area under the curve; Se, sensitivity; Sp, specificity.

Discussion

This study shows that in a primary care population presenting with knee pain, in which GPs experience diagnostic challenges, an age threshold of roughly 50 years is a strong predictor for knee osteoarthritis, patellofemoral pain, and meniscal lesions, which are the three most common diagnoses.

The high prevalence of knee osteoarthritis is consistent with the findings of another Dutch study, which showed that 70% of the patients referred to an outpatient orthopaedic clinic with atraumatic knee pain were diagnosed with knee osteoarthritis.¹ Prior research also showed that patellofemoral pain and (degenerative) meniscal lesions are less commonly diagnosed than osteoarthritis.^1,5

The AUC in an ROC analysis indicates how good the predictive power of a model is in distinguishing between patients with and without a specific condition. An AUC of 1.0 means the model has perfect discrimination, while an AUC of 0.5 suggests the model’s predictive power is no better than random chance. In our study, we observe an AUC of 0.90 for osteoarthritis of the knee, and an AUC of 0.87 for patellofemoral pain, indicating the high predictive power of age for both these conditions.⁹

The probability of having knee osteoarthritis increases with age, as reflected in an increase in the PPV per 5 years of life. For example, patients aged 50 years or older have a 91% chance of being diagnosed with knee osteoarthritis, while this chance increases to 95% in patients aged 65 years or older. We observe the opposite for patellofemoral pain and meniscal lesion, where an age of 50 years or older almost certainly rules out both conditions as reflected by an NPV of 99% and 98%, respectively. The age threshold for osteoarthritis (49.5 years) is consistent with the American College of Rheumatology clinical osteoarthritis criterion for age of ≥50 years,^5,10 while patellofemoral pain and meniscal lesions typically affect younger people.⁵ Meniscal lesions typically present with knee pain localised to the joint line and an accompanying effusion, and are usually caused by twisting injuries or degenerative changes.⁵ In contrast, patellofemoral pain is primarily associated with anterior knee pain, particularly during activities like squatting, running, or climbing stairs, and is often linked to overuse or biomechanical imbalances.⁵ These differences highlight distinct etiologies and symptomatology between patellofemoral pain and meniscal lesions.

While our results should be interpreted with some caution because of the single-centred design, we emphasise the strength of our setting. Specifically, the setting of a clinic for low-complexity cases like ours fits well with the population seen in general practice. Moreover, the frequency of the encountered diagnosis is in line with prior research, implying that our results seem generalisable. Our results suggest that age alone could be a simple diagnostic tool to assist the GP in the clinical decision making process. Nonetheless, given knee pain is a multifaceted condition, other factors may be important in predicting diagnosis. Only including age may therefore be considered a limitation of this study. Therefore, we plan to develop a multifactorial machine-learning-based prediction model for diagnosing knee conditions in those patients where the GP faces diagnostic challenges.

In patients presenting with knee pain, for whom the GP faces diagnostic challenges, GPs should consider an age threshold of roughly 50 years as a strong predictor for the three most common diagnoses: knee osteoarthritis, patellofemoral pain, and meniscal lesions. This helps the GP select the most appropriate management plan and also informs the prognosis.