SPOT: An Early Detection Tool for Predicting Pain Trajectories Following Intra‐Articular Injections

ABSTRACT

The purpose of this study was to identify 12‐month postinjection knee osteoarthritis (KOA) pain trajectories and develop an early prediction model using patient‐reported pain scores. Data were derived from a multi‐site, single‐blind randomized controlled trial of intra‐articular injections for KOA. The primary analysis focused on participants assigned to autologous bone marrow aspirate concentrate, umbilical cord tissue‐derived mesenchymal stromal cells, or stromal vascular fraction. After data‐quality screening, the primary trajectory cohort included 317 participants and the prediction cohort included 295 participants with Screening and Month‐3 KOOS‐12 Pain scores. Latent class mixed‐effects modeling identified a two‐class solution selected for clinical interpretability: 66.6% of patients followed an Improved Pain trajectory and 33.4% followed a Persistent Pain trajectory. Although a three‐class model had slightly lower BIC, the additional class included only 2.5% of participants and was not retained for the clinical prediction tool. A logistic regression model using KOOS‐12 Pain at Screening and 3‐month change predicted trajectory membership with AUC = 0.785 in training and AUC = 0.792 in the held‐out test set. A training‐selected threshold of 0.656 yielded held‐out test sensitivity = 0.790, specificity = 0.692, PPV = 0.860, NPV = 0.581, and accuracy = 0.761. The resultant model was implemented as the Symptom Prediction for Outcomes of Treatment (SPOT) web calculator. Clinical Significance: The SPOT web calculator only requires KOOS‐12 Pain data from Screening and Month 3 to classify patients as likely to follow an Improved Pain vs. Persistent Pain trajectory after orthobiologic KOA treatment.

Trial Registration: ClinicalTrials.gov identifier: NCT03818737

1.

Knee osteoarthritis (KOA) results in cartilage degradation, substantial joint pain, loss of functional mobility, and a significant financial burden [1]. Approximately 4% of adults are treated for KOA, resulting in an estimated $15 billion in healthcare costs [2]. The sustained personal health burden and the increased national healthcare costs are primarily due to the lack of an effective condition‐modifying treatment—one that can slow the progression of the condition while providing long‐term symptomatic relief from knee pain. Among available treatment options, intra‐articular injections show significant promise for providing pain relief from KOA [1]. Intra‐articular injection therapies, such as umbilical cord tissue‐derived mesenchymal stromal cells (UCT), bone marrow aspirate concentrate (BMAC), and stromal vascular fraction (SVF), have shown promising preliminary efficacy in reducing pain and improving knee function in patients with KOA [3, 4, 5]. Following injections, pain relief is not consistently effective long‐term, and varies widely across patients and injections; decisions to continue, switch, or stop treatment therefore remain uncertain.

Longitudinal KOA studies show significant heterogeneity in pain trajectories over time [6, 7, 8, 9]. Across cohorts, investigators have identified multiple pain trajectories (e.g., stable/mild pain, improving pain, and persistently high or worsening pain) even when patients receive the same or similar treatment [7, 10]. Grouping patients by their pain trajectories over time, rather than at a single time point, can reveal clinically meaningful differences in prognosis and treatment response. From a clinical standpoint, such trajectory information could help differentiate patients who are likely to maintain pain relief from those whose symptoms will remain high or rebound after an initial response. However, previous studies have not evaluated pain trajectories in patients receiving intra‐articular injections.

Variability in patients' treatment response from intra‐articular injections is multifactorial and remains poorly understood. Patient‐specific factors such as joint anatomy and alignment, muscle strength, hormonal influences, obesity, and genetic predispositions likely contribute to variable outcomes with intra‐articular injections [1, 11, 12]. There are currently no best‐practice clinical guidelines to guide the planning of intra‐articular injection treatments based on individual patient characteristics [13]. Clinically accessible tools that can identify patients' likelihood of pain relief early in treatment would support clinical decision‐making, enabling early informed treatment decisions. When such tools rely on patient‐reported outcome measures that are already collected as part of standard care, they minimize additional burden while providing structured risk information. An early prediction tool would reduce the financial burden and improve patient pain and function.

There remains an unmet need to identify which intra‐articular injection therapies are most effective for individual patients affected by KOA. Clinically accessible tools that can flag early in care whether a patient is on a favorable or unfavorable path of pain recovery would enable the timely delivery of alternative or additional interventions and discontinuation of ineffective treatments. Identification of patients' pain trajectories could also help clinicians counsel patients more about expected recovery. Accordingly, the present study had two related aims. First, we sought to derive and characterize clinically meaningful pain‐recovery trajectories in patients with KOA who received orthobiologic intra‐articular injections using latent class mixed‐effects modeling. Second, we aimed to develop a prediction model and online calculator that uses early KOOS‐12 Pain scores to classify individual patients into those statistically derived pain‐trajectory groups.

2. Methods

2.1. Study Institution and Patients

Data were derived from a single‐blind randomized controlled trial of intra‐articular injections for KOA (ClinicalTrials.gov Identifier: NCT03818737). In the parent trial, 480 participants were randomized and 475 were included in the intention‐to‐treat sample across four parallel groups: autologous bone marrow aspirate concentrate (BMAC; n = 118), autologous stromal vascular fraction (SVF; n = 119), allogeneic umbilical cord tissue–derived mesenchymal stromal cells (UCT; n = 118), and corticosteroid injection (CSI; n = 120) [14]. Informed consent was obtained prior to commencing the study and was approved by the institutional review board of Emory University.

For this secondary prognostic analysis (level of evidence II), we excluded the CSI arm because its effect profile is rapid and typically short‐lived relative to orthobiologic injections [14, 15, 16, 17], and we restricted analyses to participants assigned to BMAC, SVF, or UCT (orthobiologic intention‐to‐treat n = 355). After prespecified data‐quality screening, the primary trajectory‐derivation cohort included 317 orthobiologic‐treatment participants with at least one post‐treatment KOOS‐12 Pain score between Months 1 and 12. A stricter cohort requiring at least three post‐treatment KOOS‐12 Pain assessments was retained as a sensitivity/quality‐assurance dataset. For prediction modeling, we further required the availability of Screening and Month‐3 KOOS‐12 Pain to compute Δ3; 295 participants met these requirements and were split into a training set (n = 207) and a held‐out test set (n = 88).

2.2. Pain Measurement

Patients completed the 12‐item Knee injury and Osteoarthritis Outcome Score (KOOS) [18, 19, 20]. KOOS‐12 Pain was assessed at Screening and Months 1, 3, 6, 9, and 12 after treatment. KOOS‐12 Pain scores were derived from the nine KOOS Pain item responses and scored on a 0–100 scale, with higher scores indicating less pain. To facilitate scoring in routine practice, clinicians can use the available online KOOS‐12 from Orthotoolkit, which reports the Pain subscale score used in this study (see: https://orthotoolkit.com/koos-12/).

2.3. Statistical Analyses

All analyses were performed using RStudio [21]. Data management was conducted using dplyr, tidyr, readxl, and stringr. Latent class mixed‐effects models were fit using lcmm::hlme() and lcmm::gridsearch(). Receiver operating characteristic curves, area under the curve estimates, and DeLong model comparisons were conducted using pROC. Calibration was evaluated using the Hosmer–Lemeshow test from ResourceSelection, and figures were produced using ggplot2. Trajectory models [22] used maximum likelihood estimation with available post‐treatment KOOS‐12 Pain observations. The primary trajectory model specified fixed = KOOS‐12 Pain ~ timepoint, mixture = ~ timepoint, random = ~1, subject = SUBJID_NUM, and nwg = TRUE for multi‐class models. Package versions and full session information are provided in the reproducible analysis output.

2.3.1. Data Quality Assurance

To ensure data quality and remove careless responders, we used three standard tests applied to KOOS Pain item responses: long‐string detection (> =12 identical item responses) [23, 24], even–odd split‐half reliability (r < 0.20) [25, 26], and Mahalanobis distance outlier detection [27, 28]. Participants were removed if any visit failed two or more of these checks. This participant‐level rule removed 13 participants and 64 visits, yielding 467 participants and 2346 KOOS Pain visits after quality control.

2.3.2. Aim 1: Trajectory Derivation Using Latent Class Mixed‐Effects Modeling

We applied latent class mixed‐effects modeling to classify patients based on post‐treatment KOOS‐12 Pain trajectories over the first year following orthobiologic treatment. We compared models with 1 to 5 classes. The primary model was fit using lcmm::hlme() with a linear effect of time, class‐specific intercepts and slopes (mixture = ~timepoint), a random intercept (random = ~1), subject‐level clustering by SUBJID_NUM, and proportional class‐specific residual variance (nwg = TRUE). Thus, unlike a strict LCGA specification, the final model allowed within‐class heterogeneity in baseline pain levels through the random intercept while retaining class‐specific fixed effects for the pain trajectory. Each model was estimated using grid‐search random starts from the one‐class solution. Final selection of the number of classes was based on fit indices, posterior‐probability diagnostics, convergence, minimum class size, parsimony, and clinical interpretability. Bayesian Information Criterion (BIC), Akaike's information criterion (AIC), and Consistent AIC (CAIC) were used to evaluate model fit [29]. The posterior‐probability diagnostics included the entropy R [2], class‐specific mean posterior probabilities, the proportion of participants with posterior probabilities >= 0.80, and overall classification error.

The primary trajectory model used KOOS‐12 Pain scores from Months 1, 3, 6, 9, and 12 post‐treatment. Screening was excluded from trajectory derivation so that the latent classes reflected post‐treatment response patterns rather than baseline symptom severity. Time was modeled as a continuous linear predictor across post‐treatment months. In the final two‐class solution, we labeled the trajectories “Improved Pain” and “Persistent Pain” to reflect their relative longitudinal KOOS‐12 Pain patterns over 12 months (higher scores indicate less pain).

2.3.3. Aim 2: Prediction and Model Development

To develop the early‐flagging tool for pain response, we tested logistic regression models to predict trajectory class membership. The final clinically deployable model used two routinely available inputs: Screening KOOS‐12 Pain and Δ3 (Month 3 – Screening). Because Month‐3 KOOS‐12 Pain was included in both the trajectory model and the prediction model, this model should be interpreted as an early trajectory‐classification model rather than as prediction of an entirely independent future endpoint.

The prediction model was trained on 70% of the eligible prediction cohort and evaluated on the remaining 30%. The training data were used to estimate model parameters and select an operating threshold for the predicted probability of Improved Pain membership. The threshold was selected in the training set to balance sensitivity and specificity and was then applied without modification to the held‐out test set. Model performance was evaluated using AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, Brier score, and Hosmer–Lemeshow calibration testing. Differences in AUCs between candidate models were compared using the DeLong test [30].

2.3.4. Tool Development

To maximize clinical utility, we implemented the final logistic regression model as an online calculation tool, the Symptom Prediction for Outcomes of Treatment (SPOT) tool. The SPOT tool requires two inputs: the patient's KOOS‐12 Pain scores at Screening and Month 3, from which the change score (Δ3 = Month 3 – Screening) is computed internally. For each patient, the calculator returns the predicted probability of Improved Pain membership and a binary classification based on the training‐selected probability threshold (see Results for model selection and performance). Patients below the threshold are classified as likely to follow the Persistent Pain trajectory.

3. Results

3.1. Data Quality Assurance

Of 475 participants in the parent trial intention‐to‐treat sample, we excluded the corticosteroid injection arm (n = 120) and restricted analyses to the orthobiologic arms (BMAC, SVF, UCT; n = 355). KOOS‐12 Pain scores were derived for 480 randomized participants contributing 2410 KOOS Pain visits before quality control. The participant‐level quality‐control rule removed 13 participants and 64 visits, yielding 467 participants and 2346 KOOS Pain visits after quality control. The primary orthobiologic trajectory cohort included 317 participants with at least one post‐treatment KOOS‐12 Pain score between Months 1 and 12. For prediction modeling, 295 participants had both Screening and Month‐3 KOOS‐12 Pain available to compute Δ3 (training n = 207; test n = 88).

3.2. Aim 1: Trajectory Derivation Using Latent Class Mixed‐Effects Modeling

Table 1 presents the fit indices for the 1‐ to 5‐class trajectory models. The three‐class model had the lowest BIC (11072.30) and higher entropy (R2 = 0.695) than the two‐class model. However, the additional class contained only eight participants (2.5% of the analytic trajectory cohort), which was too small to support a clinically deployable prediction tool. The two‐class model had slightly higher BIC (11078.44) and lower entropy (R2 = 0.530), but it separated the sample into two clinically interpretable and adequately sized classes: Improved Pain (n = 211, 66.6%) and Persistent Pain (n = 106, 33.4%). Although a formal minimum class‐size threshold was not prespecified, class size and clinical interpretability were part of the final model‐selection criteria. We therefore retained the two‐class solution for parsimony, interpretability, and clinical deployment while acknowledging the statistical fit advantage of the three‐class solution.

Table 1.

Measures of model fit and classification accuracy for 1‐ to 5‐class post‐treatment KOOS‐12 Pain trajectory models.

G	No. parameters	BIC	AIC	CAIC	Classification error	Entropy R2	Smallest class n (%)	Decision
1	4	11,132.66	11,117.62	11,136.66	—	—	—	Rejected: poorer fit
2	8	11,078.44	11,048.37	11,086.44	0.145	0.530	106 (33.4%)	Selected for clinical interpretability
3	12	11,072.30	11,027.19	11,084.30	0.140	0.695	8 (2.5%)	Rejected: very small third class
4	16	11,081.00	11,020.86	11,097.00	0.230	0.609	10 (3.2%)	Rejected: small class/less parsimonious
5	20	11,101.29	11,026.11	11,121.29	0.280	0.594	10 (3.2%)	Rejected: small class/less parsimonious

Posterior‐probability diagnostics for the retained two‐class model showed moderate entropy (R2 = 0.530), mean maximum posterior probability of 0.855, and estimated classification error of 0.145. Overall, 71.6% of participants had posterior probabilities >= 0.80. By class, the Improved Pain group had a mean posterior probability of 0.870 (75.4% >= 0.80), whereas the Persistent Pain group had a mean posterior probability of 0.827 (64.2% >= 0.80) (Table 2). These values indicate clinically useful but imperfect class separation and motivated cautious interpretation of the prediction model.

Table 2.

Retained two‐class trajectory solution and posterior‐probability diagnostics.

Class	n	%	Mean KOOS‐12 Pain	Mean posterior probability	% posterior probability >= 0.80
Improved pain	211	66.6	78.30	0.870	75.4
Persistent pain	106	33.4	57.02	0.827	64.2
Overall	317	100.0	—	0.855	71.6

The final two‐class model identified two statistically derived longitudinal pain‐trajectory classes. The Improved Pain class was characterized by higher KOOS‐12 Pain scores over follow‐up, indicating less pain and more favorable response. The Persistent Pain class was characterized by lower KOOS‐12 Pain scores and less favorable improvement across follow‐up. Of the 317 participants included in the primary trajectory cohort, 211 (66.6%) followed the Improved Pain trajectory and 106 (33.4%) followed the Persistent Pain trajectory (Figure 1).

Figure 1.

KOOS‐12 Pain trajectories over 12 months by statistically derived trajectory class. Thin transparent lines represent observed individual participant trajectories; thicker lines represent class‐specific mean trajectories. Higher KOOS‐12 Pain scores indicate less pain. The latent class mixed‐effects model identified two clinically interpretable classes: Improved Pain (red) and Persistent Pain (blue).

3.3. Aim 2: Prediction Model Performance and SPOT Tool

We trained logistic models to predict trajectory membership (Improved Pain = 1 vs. Persistent Pain = 0) using KOOS‐12 Pain data available at early follow‐ups. A single 70/30 split was used for the primary internal validation (training n = 207; test n = 88). On the training set, we selected a single operating probability threshold to balance sensitivity and specificity for the predicted probability of Improved Pain membership. The same threshold was then applied without modification to the held‐out test set.

Models that included Month‐6 information showed the highest discrimination but were not selected as the final model because they would require an additional 3 months before classification. The Screening + Δ3 model showed strong performance in both training and held‐out test data (Table 3; Figure 2). The fitted model was logit P(Improved Pain) = −4.3972 + 0.07319 × Screening KOOS‐12 Pain + 0.07085 × Δ3. Higher Screening KOOS‐12 Pain and greater improvement by Month 3 were associated with higher odds of membership in the Improved Pain trajectory class (Table 4). Using the training data, we selected a threshold of 0.656 on the predicted probability of Improved Pain. At this threshold, the model achieved test‐set AUC = 0.792, sensitivity = 0.790, specificity = 0.692, PPV = 0.860, NPV = 0.581, accuracy = 0.761, Brier score = 0.154, and acceptable calibration by Hosmer–Lemeshow test (p = 0.636).

Table 3.

Prediction‐model performance for Screening + Δ3 model.

Set	n	AUC	Threshold	Sensitivity	Specificity	PPV	NPV	Accuracy	Brier score	Hosmer–Lemeshow p
Training	207	0.785	0.656	0.733	0.708	0.825	0.586	0.725	0.18	—
Held‐out test	88	0.792	0.656	0.790	0.692	0.860	0.581	0.761	0.154	0.636

Figure 2.

ROC curves for the Screening + Δ3 model predicting 12‐month Improved Pain trajectory membership in the training and held‐out test sets. The threshold selected in the training set was 0.656 and was applied without modification to the held‐out test set. The diagonal dashed line represents chance‐level classification.

Table 4.

Logistic regression coefficients for the final Screening + Δ3 model predicting Improved Pain membership.

Term	Estimate	SE	OR	95% CI for OR	p value
Intercept	−4.397	0.927	0.012	0.002–0.076	< 0.001
Screening KOOS‐12 Pain	0.073	0.015	1.076	1.045–1.108	< 0.001
Δ3 (Month 3 – Screening)	0.071	0.013	1.073	1.047–1.101	< 0.001

4. Discussion

Identifying early whether patients are likely to follow a favorable or unfavorable pain‐response trajectory may support more personalized treatment decisions. This study had two related aims. First, we identified statistically derived pain‐trajectory classes across the first 12 months following orthobiologic intra‐articular injections. Results supported two clinically interpretable classes: Improved Pain with sustained gains (66.6% of participants) and Persistent Pain with lower or less sustained improvement (33.4%). Second, those empirically identified trajectories were used to inform a two‐input early classification tool. Using only KOOS‐12 Pain at Screening and 3‐month change, the model achieved AUC = 0.785 in training and AUC = 0.792 in the held‐out test set, with test‐set sensitivity = 0.790 and specificity = 0.692. Clinicians can use the free online SPOT calculator to support early identification of patients likely to remain on an Improved Pain vs. Persistent Pain trajectory.

To the authors' knowledge, this is the first study to identify KOA pain‐trajectory classes specifically following intra‐articular orthobiologic injections. Previous studies on KOA have shown substantial heterogeneity in pain trajectories over time [31]. Variations in pain trajectories range from three to five distinct groups, from stable, mild pain to persistently high or worsening pain [7, 32]. In parallel, there is significant variation in the effectiveness of orthobiologic treatments for KOA, as shown through clinical trials and documented with systematic reviews [4, 33]. Our selected two‐class model should be interpreted as a clinically deployable simplification of a more complex response distribution: although the three‐class model had the lowest BIC, the additional class was very small (2.5%), whereas the two‐class model provided two adequately sized and clinically interpretable groups aligned with practical decisions about whether to continue standard follow‐up or consider treatment adjustment.

Our second aim was to develop an early clinical prediction tool. Using KOOS‐12 Pain outcomes, we created a two‐input logistic model that, by Month 3, estimates a patient's probability of belonging to the Improved Pain vs. Persistent Pain trajectory. Implemented as the SPOT web calculator, this model allows clinicians to receive an early trajectory classification (see Figure 3). SPOT can be integrated into existing workflows without additional specialized testing. The tool is intended to support, but not replace, clinical judgment by providing quantitative risk information that can be used alongside all available medical information, patient preferences, and clinician expertise.

Figure 3.

Example outputs from the SPOT tool. SPOT estimates the probability of 12‐month Improved Pain membership using Screening KOOS‐12 Pain and Delta3 (Month 3 – Screening), then applies the training‐selected threshold to classify patients as likely Improved Pain or likely Persistent Pain. Panel A shows a Persistent Pain classification, suggesting closer follow‐up or consideration of treatment adjustment; Panel B shows an Improved Pain classification, supporting continuation of standard care and monitoring.

Although this study provides significant contributions, additional steps should be taken before widespread adoption. First, external validation should be conducted. Results were based on a single randomized trial; performance in other clinical settings and patient populations has yet to be determined. Second, the retained two‐class model had moderate entropy (R2 = 0.530) and classification error of 0.145, indicating that class membership should not be interpreted as perfectly observed. Because the prediction model used modal class assignment as the binary outcome, predictive performance may be somewhat optimistic. Future work should incorporate classification uncertainty using three‐step, posterior‐probability‐weighted, pseudo‐class, or one‐step approaches. Third, Month‐3 KOOS‐12 Pain contributed to both trajectory estimation and the early prediction model. SPOT should therefore be interpreted as early classification of the likely 12‐month trajectory pattern, not as prediction of an entirely independent future endpoint. Future studies should validate SPOT using independently derived later‐outcome trajectories and external cohorts. Finally, because the present analyses focused on orthobiologic injections, generalizability to other intra‐articular treatment modalities will require additional testing.

SPOT is designed to be deployed at a routine 3‐month follow‐up visit, using KOOS‐12 Pain scores that are already commonly collected in medical contexts. For a given patient, the clinician can enter the Screening and Month‐3 KOOS‐12 Pain scores into the web‐based calculator, review the predicted pain trajectory class, and discuss the implications with the patient. Patients classified as likely Persistent pain may warrant additional interventions (e.g., rehabilitation and weight loss), alternative injection, or surgery. SPOT has the potential to reduce repeated ineffective injections and support more informed decision‐making among clinicians.

5. Conclusions

We developed and internally validated an early‐flagging logistic model that classifies patients as likely to follow an Improving or Persistent 12‐month pain trajectory after intra‐articular orthobiologic injection. Presented as the SPOT web calculator, this model shows that routinely collected PRO data can be leveraged efficiently and effectively to predict long‐term effectiveness of treatment. Although further external validation studies should be conducted, these findings provide a foundation for leveraging PRO results to inform future KOA treatment decisions.

Author Contributions

All authors read and approved the final submitted manuscript. Contributions of each author listed after appropriate categories below (using initials). Substantial contributions to research design, or the acquisition, analysis or interpretation of data: Brittany L. Nelson, Prathap Jayaram, Scott D. Boden, Kenneth Mautner, Michael Gottschalk, Gregory D. Myer, Hicham Drissi, and Jed A. Diekfuss. Drafting the paper or revising it critically: Brittany L. Nelson, Prathap Jayaram, Kenneth Mautner, Gregory D. Myer, and Jed A. Diekfuss. Approval of the submitted and final versions: Brittany L. Nelson, Prathap Jayaram, Scott D. Boden, Kenneth Mautner, Michael Gottschalk, Gregory D. Myer, Hicham Drissi, and Jed A. Diekfuss.

Conflicts of Interest

Kenneth Mautner is a consultant for Lipogems, which is a micronized fat company, not used in this current study. Gregory D. Myer has no direct conflicts of interest related to the current investigation. His institutions have received past and/or ongoing research funding from the National Institutes of Health/National Institute of Arthritis and Musculoskeletal and Skin Diseases, the U.S. Department of Defense, the U.S. Department of Veterans Affairs (CReATE Motion Center), and the Arthritis Foundation Osteoarthritis Clinical Trial Network. Dr. Myer has also received industry‐sponsored research support paid to his institution for research related to injury prevention, sports performance, rehabilitation, imaging, and surgical interventions. Current support includes funding from Arthrex Inc. for evaluation of ACL surgical techniques; GE HealthCare for MRI sequence development; the National Basketball Association for validation of biomechanical models; and Medacta International for joint implant research. His research programs also benefit from philanthropic support from the Arthur M. Blank Family Foundation and the Joe and Clara Tsai Foundation through the Wu Tsai Human Performance Alliance to advance human performance research, discovery, and dissemination. Dr. Myer receives author royalties from Human Kinetics and Wolters Kluwer and is an inventor of biofeedback technologies (U.S. Patent No. 11,350,854 B2, “Augmented and Virtual Reality for Sport Performance and Injury Prevention Application,” issued June 7, 2022, with associated software copyright), which have generated licensing royalties. Jed A. Diekfuss receives author royalties from Kendall Hunt Publishing Company and research funding from the U.S. Department of Veterans Affairs (CReATE Motion Center), GE HealthCare, and Arthrex Inc. for studies focused on ACL injury and reconstruction, post‐traumatic osteoarthritis, and related clinical outcomes. Dr. Diekfuss' disclosed activities are not directly related to the current investigation and do not represent a conflict of interest with this study.

Spot Tool Web Calculator Access Link

https://survey.qualtrics.emory.edu/jfe/form/SV_cGc5V2Muut68WUe.

Nelson B. L., Jayaram P., Boden S. D., et al., “SPOT: An Early Detection Tool for Predicting Pain Trajectories Following Intra‐Articular Injections,” Journal of Orthopaedic Research® 44 (2026): e70270. 10.1002/jor.70270.

Data Availability Statement

Data may be made available via a two‐stage approach. First, the investigator(s) requesting data will need to submit a proposal to the authorship team that delineates the purpose of the request, the data elements requested, the proposed analyses, and plan for dissemination of new findings. Data requests should be reasonable, not place an undue burden on the study team, and the available data that can be provided is restricted to the specific data that supports the study findings of the published manuscript, only. The authors have the right to not approve a request that seeks to answer new research questions or falls outside the scope of replication‐based purposes. Second, if the proposal is approved by the authorship team, data will only be made available after meeting all regulatory requirements, institutional approvals have been filed, and signed, multi‐site data use agreements are in place. The authorship team also intends to share this data publicly in the future, after all manuscripts have been published, such that the above two‐stage approach will not be necessary.