This multisite, single-blind randomized controlled trial analyzed patient-reported pain to identify 12-month postinjection pain trajectories in knee osteoarthritis (KOA) and to develop an early prediction model. The focus was on participants who received orthobiologic intra-articular injections: autologous bone marrow aspirate concentrate, umbilical cord tissue-derived mesenchymal stromal cells, or stromal vascular fraction. The objective was to use routinely collected KOOS-12 Pain scores to predict which patients would follow an Improved versus Persistent pain trajectory over 12 months.
Data were derived from the randomized controlled trial noted above. After data-quality screening, two analytic cohorts were defined:
Latent class mixed-effects modeling was applied to longitudinal KOOS-12 Pain data to identify discrete pain trajectories. For the prediction task, a logistic regression model used Screening KOOS-12 Pain and the change at Month 3 as predictors of trajectory membership.
Latent class mixed-effects modeling identified a two-class solution chosen for clinical interpretability. The two classes represented distinct 12-month pain courses:
Although a three-class model produced a slightly lower Bayesian information criterion (BIC), the additional class comprised only 2.5% of participants and was not retained for the clinical prediction tool, based on limited clinical utility of that very small subgroup.
A logistic regression model was developed to predict trajectory membership using only two inputs readily available in routine follow-up: the KOOS-12 Pain score at Screening and the change in KOOS-12 Pain from Screening to Month 3. The model was trained and evaluated with a held-out test set to assess generalizability.
Model discrimination in the training set was reported as area under the receiver operating characteristic curve (AUC) = 0.785. Performance in the held-out test set was similar, with AUC = 0.792.
A probability threshold of 0.656 was selected based on training-set tuning and then applied to the held-out test set. With that threshold, held-out test metrics were:
These metrics indicate the model was better at ruling in the Improved Pain trajectory (high PPV) than ruling out Persistent Pain (moderate NPV) in the held-out data.
The final logistic regression model was implemented as the Symptom Prediction for Outcomes of Treatment (SPOT) web calculator. SPOT requires only the Screening KOOS-12 Pain score and the Month-3 KOOS-12 Pain score (or the change from Screening to Month 3) to classify patients as likely to follow an Improved Pain versus Persistent Pain trajectory after orthobiologic KOA treatment.
The authors present SPOT as a simple, patient-reported outcome–based tool that can be applied early (by Month 3) to estimate 12-month pain trajectory following orthobiologic intra-articular injections for KOA. The tool leverages the KOOS-12 Pain measure and a parsimonious prediction model to support clinical classification of likely outcomes, potentially informing follow-up plans or shared decision-making. The report emphasizes that the two-class trajectory solution was selected for clinical interpretability.
The randomized controlled trial underlying this analysis is registered at ClinicalTrials.gov under identifier NCT03818737. Reported keywords include KOOS-12, knee osteoarthritis, latent class mixed-effects modeling, patient-reported outcomes, and prediction model.
The abstract reports cohort sizes, modeling approach, selected trajectory solution, model predictors, discrimination metrics, a chosen probability threshold, and held-out test performance. Further methodological details, full model coefficients, external validation, and limitations beyond what is summarized in the abstract were not reported in the provided source text and would require consultation of the full article for complete appraisal.