Early Prediction of Scar Outcomes: A Prospective Study on the Predictive Value of 3‐Month Scar Assessments for 12‐Month Results

ABSTRACT

Scar outcomes are traditionally evaluated at 12 months post‐surgery, yet early prediction of long‐term scar characteristics may facilitate timely interventions, enhance treatment strategies, shorten clinical trial durations, and improve patient outcomes. This study evaluates whether 3‐month scar assessments reliably predict 12‐month outcomes using validated scar assessment scales and interrater reliability measures. Two surgeons evaluated a prospective cohort of 40 surgical patients using standardised scar assessment tools. Logistic regression assessed the predictive value of 3‐month classifications for 12‐month binary outcomes (‘good’ vs. ‘poor’ scar quality), while linear regression evaluated continuous scar scores. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values quantified predictive accuracy. Scars classified as ‘good’ at 3 months were significantly more likely to remain ‘good’ at 12 months (p < 0.001), with strong predictive performance (AUC = 0.763). Linear regression demonstrated significant predictive accuracy for width, height, and overall opinion. However, vascularity predictions were weak, and depth was not significantly associated with long‐term outcomes. Interrater reliability was moderate to good across measures, with intraclass correlation coefficients values ranging from 0.215 (depth) to 0.714 (width), and Cohen's Kappa for binary scar classifications of 0.505 (p < 0.001). Early 3‐month scar assessments reliably predict 12‐month outcomes for width, height, and overall opinion, while vascularity and depth remain less predictable. Scars rated as ‘good’ at 3 months are unlikely to worsen, while ‘poor’ scars rarely improve without intervention. These findings highlight the utility of early scar evaluations in guiding postoperative management and patient counselling.

Key Points

• Three‐month scar assessments can meaningfully predict 12‐month outcomes, though predictive value varies considerably by feature.

• Scar width was the strongest and most reliable early predictor, with the highest regression coefficient, explained variance, and inter‐rater agreement of all assessed features.

• The binary “good/poor” gestalt classification achieved 100% sensitivity and negative predictive value, meaning scars judged poor at 3 months reliably remained poor at 12 months.

• Vascularity and depth showed weak predictive associations and poor inter‐rater reliability, cautioning against early intervention decisions based on these parameters alone.

• The cohort's limited racial and ethnic diversity underscores the need for larger, more representative studies given the elevated risk of pathological scarring in patients of color.

1. Introduction

Scarring represents an unavoidable consequence of wound healing after surgery or injury, with its ultimate appearance exerting substantial effects on patients' physical function and psychological well‐being. Wound healing progresses through three overlapping phases: inflammation, proliferation and remodelling [1, 2, 3, 4, 5, 6]. Remodelling can last up to 12–18 months, during which collagen maturation and vascular regression contribute to the eventual appearance of the scar [6, 7]. This prolonged course complicates efforts to predict final outcomes early in the healing process.

While the cellular mechanisms of scar maturation are increasingly studied, the temporal dynamics of aesthetic outcomes remain less understood [8]. Aberrant maturation can lead to pathological scars with excessive thickness, elevation, and persistent symptoms such as pain and pruritus [9]. Beyond physical discomfort, these scars can cause significant psychological distress, including internalised stigmatisation and reduced quality of life [10, 11]. Even nonpathological outcomes like dyspigmentation or widening may impair satisfaction and psychosocial well‐being [12, 13].

To date, most clinical studies evaluating antiscarring interventions have relied on assessments conducted at later stages of healing, typically 12 months or beyond, when scars are presumed to have reached ‘final’ maturation [14, 15]. Limited research exists on the predictive value of earlier evaluations, particularly at 3 months when scars are still maturing but already exhibit measurable characteristics such as vascularity, height, width, depth, and pigmentation [16]. The early identification of stable, predictive metrics could provide valuable guidance for postoperative care and enable timely therapeutic interventions. Furthermore, in the context of emerging antiscarring drug development, validated early surrogate endpoints could substantially reduce the duration and cost of clinical trials.

This prospective study aims to evaluate whether scar characteristics assessed at 3 months postoperatively can reliably predict 12‐month outcomes. These findings have the potential to inform clinical decision‐making, improve scar management protocols, facilitate therapeutic trials, and enhance patient satisfaction through early intervention and realistic expectation setting.

2. Methods

2.1. Trial Design/Setting

Our study was conducted in the Division of Plastic and Reconstructive Surgery at an academic institution. The surveys were administered to patients who had undergone surgery by two board‐certified plastic surgeons. Recruitment began in November of 2021, and data collection continued until August of 2024. This study was approved by the site's Institutional Review Board.

2.2. Study Participants

Patients coming into the clinic for standard postoperative visits were approached regarding involvement in the present research study. None of the patients in the cohort were smokers. Patients 18 years of age or older presenting with visible scars at their 3‐month postoperative visit were eligible. Participants who underwent minimally invasive procedures that did not result in visible scars were excluded from the study. Exclusion criteria also included any medical or psychiatric condition that would complicate patient compliance or interfere with the interpretation of study results. Eligibility to participate was determined by the judgement of the Principal Investigator. All patients provided written informed consent. No scar‐specific preventive interventions were administered between 3 and 12 months postoperatively.

2.3. Data Collection

The cohort included a heterogeneous range of reconstructive and aesthetic procedures to reflect real‐world clinical practice. Surgical procedures included breast reduction, abdominoplasty, Mohs reconstruction, rhinoplasty, implant exchange, breast revision, and other soft‐tissue operations. For consistency, all assessments focused on a single representative scar region per patient. Surgical variables such as incision location, depth, closure technique, and tension vectors were not controlled, as the study's aim was to evaluate the clinical utility of early scar assessments across varied operative contexts rather than within a narrowly defined procedural subgroup.

Two board‐certified plastic surgeons independently evaluated each scar. The raters have decades of experience in plastic surgery and scar management; they have both used and are familiar with the POSAS scale [17]. Observer assessments included: (1) qualitative characterisation of skin tone and scar features (pigmentation, vascularity, width, height, and depth); (2) the POSAS observer scale and (3) a binary overall classification of scars as ‘good’ (flat, pale and well‐blended) or ‘poor’ (raised, widened, vascular or dyspigmented). This binary outcome was designed to capture clinician gestalt, reflecting how surgeons integrate multiple visual features into actionable judgements during routine practice (Figure 1).

FIGURE 1.

Representative ‘good’ and ‘poor’ scars at timepoints of 3 and 12 months postoperative.

Surgeons also recorded both current and predicted 12‐month scar characteristics at the 3‐month visit using standardised visual reference scales (vascularity and width: 1–10; height and depth: 1–6) (Figure 2). These visuals act as a calibration aid to standardise rater interpretation of scar depth and height. At the 12‐month visit, surgeons reassessed scars using the same scales without making predictions. Surgeons were blinded to their prior ratings. For consistency, all evaluations focused on a single scar region (e.g., one breast in patients with bilateral reductions).

FIGURE 2.

Visual aids incorporated into the scales of the questionnaire.

2.4. Statistical Analysis

All analyses were performed in R. Descriptive statistics (means, standard deviations) summarised scar characteristics at 3 and 12 months. Predictive relationships for continuous outcomes (depth, height, width, vascularity, overall opinion) were assessed with linear regression and Pearson correlations; scatterplots with regression lines illustrated these associations. This study was designed as an exploratory, hypothesis‐generating investigation; therefore, no formal a priori sample size calculation was performed.

Binary outcomes (‘good’ vs. ‘poor’) were evaluated with logistic regression, with model performance reported using sensitivity, specificity, confusion matrices, and area under the ROC curve (AUC). Categorical outcomes (e.g., pigmentation) were analysed with matched‐pair tests, and McNemar's test assessed changes over time.

Interrater reliability was assessed as a core component of feasibility analysis. Intraclass correlation coefficients (ICC[A,2]) were calculated for continuous variables, and Cohen's Kappa was used for binary outcomes. ICC values were interpreted as poor (< 0.40), moderate (0.40–0.59), good (0.60–0.74) or excellent (> 0.75). Statistical significance was set at p < 0.05, and results were reported with confidence intervals and R ² where appropriate.

3. Results

3.1. Study Population

Among the 40 participants, six identified as Hispanic, 11 identified as African American, one as Asian, and 22 identified as Caucasian. The cohort comprised 38 women and two men, with racial and ethnic diversity reflective of the clinic population. Twenty‐six participants underwent breast‐related surgeries, with 21 undergoing a bilateral breast reduction. The other procedures involved included abdominoplasty, Mohs closure, lipoma removal, rhinoplasty, breast revision, implant exchange, and breast wound debridement (Table 1).

TABLE 1.

Demographics.

Demographics	Participants
Race
Caucasian	22
African‐American	11
Hispanic	6
Asian	1
Age
18–40	16
41–65	16
65+	8
Scar location
Breast	26
Face	9
Abdomen	3
Back	2

3.2. Good Versus Poor

Logistic regression demonstrated that 3‐month scar classification provided meaningful predictive information for 12‐month outcomes. Overall classification accuracy was 77.5%, with perfect sensitivity and negative predictive value (NPV), indicating that scars classified as poor at 3 months consistently remained poor at 12 months. Specificity was lower, reflecting a tendency to overpredict favourable outcomes. Sensitivity was 100.0%, meaning that all scars classified as ‘good’ at 3 months and predicted to remain ‘good’ were correctly identified at 12 months. However, specificity was 52.6%, indicating some overprediction of good outcomes. The positive predictive value (PPV) was 70.0%, while the NPV was 100%.

A receiver operating characteristic (ROC) curve was generated to assess the model's ability to differentiate between good and poor scars, yielding an area under the curve (AUC) of 0.76, indicating moderate predictive performance. A McNemar's test (p = 0.0077) showed a significant difference in misclassification patterns, particularly in overpredicting good scar outcomes. Interrater agreement for binary classification was moderate (κ = 0.51), confirming reasonable consistency while highlighting subjectivity in gestalt‐based assessment. The prediction system was generally reliable, with some subjectivity remaining in scar assessments and potentially contributing to classification errors (Figure 3).

FIGURE 3.

Receiver operating characteristic (ROC) curve for logistic regression model predicting binary scar quality (‘good’ vs. ‘bad’) at 12 months based on early scar metrics. The area under the curve (AUC) quantifies model performance in distinguishing between outcome groups.

3.3. Overall Opinion

A linear regression model showed a statistically significant positive association between early predictions and actual long‐term assessments (β = 0.78, p < 0.001). The intercept was not significant (β = 1.05, p = 0.12), suggesting that the observed trend was primarily driven by the relationship between the predictor and outcome rather than an inherent baseline effect. The model explained 30.3% of the variance in 12‐month overall opinion scores (R ² = 0.31, adjusted R ² = 0.28, F(1,38) = 16.48, p < 0.001), suggesting that while early scar predictions provide a meaningful estimate of long‐term scar quality, other factors contribute to variability in outcomes. The significant regression coefficient supports the clinical relevance of early predictions in forecasting long‐term scar severity.

The ICC(A,2) = 0.59 (95% CI: 0.36–0.73, p < 0.001) indicates moderate agreement between raters. The F‐test for reliability (F(79,79.3) = 2.45, p < 0.001) confirms that the observed agreement was significantly greater than chance. These findings suggest that early predictions of overall opinion ratings at 3 months are moderately predictive of 12‐month outcomes, with each 1‐point increase in the predicted 3‐month rating associated with a 0.78‐point increase at 12 months (Figure 4).

FIGURE 4.

Prediction of 12‐month overall opinion from 3‐month scores. Scatter plot showing the linear relationship between 3 and 12‐month patients' overall opinion of scar quality (0 = normal skin, 10 = worst imaginable). A strong predictive relationship was observed. The shaded area around the regression line represents the 95% confidence interval.

3.4. Vascularity

A linear regression model identified a statistically significant but weak positive association between early predictions and long‐term vascularity (β = 0.26, p = 0.048). However, the low regression coefficient indicates that the relationship was relatively weak. The model intercept was significant (β = 0.76, p < 0.001), indicating that even in cases where 3‐month vascularity predictions were minimal, some degree of vascularity persisted at 12 months. The model explained only 4.9% of the variance in 12‐month vascularity outcomes (R ² = 0.049, adjusted R ² = 0.037, F(1,78) = 4.04, p = 0.048), suggesting that vascularity at 3 months is a poor predictor of long‐term vascularity. The residual standard error of 1.41 suggests substantial variability in individual vascularity outcomes.

The ICC(A,2) = 0.36 (95% CI: 0.001–0.59, p = 0.025) suggests poor to fair agreement between raters in assessing vascularity. The F‐test for reliability (F(79,79) = 1.56, p = 0.025) confirmed that the observed agreement was statistically significant but relatively low. These findings indicate that early vascularity assessments provide minimal predictive value for long‐term outcomes. The weak correlation and low explained variance suggest that vascular changes are highly dynamic and influenced by multiple factors beyond initial scar assessments (Figure 5).

FIGURE 5.

Prediction of 12‐month vascularity from 3‐month assessments. Linear regression plot illustrating the association between vascularity scores at 3 and 12 months. Although a positive trend was observed, the correlation was weak. The shaded area represents the 95% confidence interval of the regression line.

3.5. Width

A linear regression model demonstrated a significant positive association between early and long‐term width assessments (β = 0.80, p < 0.001). The intercept (β = 1.44, p = 0.001) was also significant, indicating that even at a baseline width of zero at 3 months, scars were likely to have some measurable width at 12 months. The model explained 37.1% of the variance in 12‐month width outcomes (R ² = 0.37, adjusted R ² = 0.36, F(1,78) = 46.01, p < 0.001). The residual standard error of 2.12 reflects some degree of variability in individual predictions.

The ICC(A,2) = 0.71 (95% CI: 0.52–0.83, p < 0.001) indicated substantial agreement between raters. The F‐test for reliability (F(79,43.8) = 3.85, p < 0.001) confirmed that the observed agreement was significantly greater than chance. These findings demonstrate that each 1‐unit increase in 3‐month width is associated with a 0.80‐unit increase at 12 months (Figure 6).

FIGURE 6.

Prediction of 12‐month width from 3‐month width assessments. Scatter plot demonstrating a strong linear correlation between 3 and 12‐month scar width measurements. Early width reliably predicted later outcomes. Shaded area reflects the 95% confidence interval.

3.6. Height

A linear regression model revealed a statistically significant positive association between early and long‐term height assessments (β = 0.59, p < 0.001). The model intercept was also significant (β = 0.64, p = 0.003). The model explained 18.8% of the variance in 12‐month height scores (R ² = 0.19, adjusted R ² = 0.18, F(1,78) = 18.04, p < 0.001). The residual standard error of 0.81 indicates moderate variability in individual predictions (Figure 7). The ICC(A,2) was 0.59 (95% CI: 0.36–0.74, p < 0.001), indicating moderate agreement between raters. The F‐test for reliability (F(79,79.7) = 2.41, p < 0.001) confirmed that the observed agreement was significantly greater than chance.

FIGURE 7.

Prediction of 12‐month height from 3‐month height assessments. This plot shows a moderate linear association between early and late scar height. Early height measurements were moderately predictive of 12‐month values. The shaded area indicates the 95% confidence interval for the regression model.

3.7. Depth

A linear regression model did not demonstrate a statistically significant association between the early and long‐term assessments of scar depth (β = 0.074, p = 0.17). The intercept was significant (β = 1.22, p < 0.001), indicating that scars tended to have a baseline depth independent of early predictions. The model accounted for only 2.4% of the variance in 12‐month depth scores (R ² = 0.0239, adjusted R ² = 0.011, F(1,78) = 1.91, p = 0.17). The residual standard error of 0.59 suggests moderate variability in individual predictions. The ICC(A,2) = 0.215 (95% CI: −0.226 to 0.498, p = 0.143) indicates poor agreement between raters. The F‐test for reliability (F(79,79.3) = 1.27, p = 0.143) further confirmed that the observed agreement was not significantly greater than chance (Figure 8).

FIGURE 8.

Prediction of 12‐month depth from 3‐month depth assessments. Linear regression plot showing the relationship between depth at 3 and 12 months. The shaded area indicates the 95% confidence interval for the regression model.

3.8. Pigmentation

A matched pairs analysis was performed to assess the agreement between pigmentation scores predicted at 3 months and those recorded at 12 months. Out of a total of 40 observations, 35 pairs (87.5%) were matched, indicating consistency in pigmentation assessments over time. However, five pairs (12.5%) were mismatched, reflecting a degree of variability between the predicted and 12‐month evaluations.

4. Discussion

This study demonstrates that select scar characteristics assessed at approximately 3 months postoperatively provide meaningful insight into 12‐month scar quality, while other features remain dynamic and less predictable during early healing. Importantly, this work represents an exploratory step toward predictive frameworks rather than a finalised prediction model.

The heterogeneity of surgical procedures included in this cohort reflects real‐world reconstructive practice and was retained to evaluate feasibility across varied operative contexts. While this heterogeneity and the limited sample size likely attenuated predictive performance for some outcomes, they enhance external validity and mirror clinical settings in which early scar assessment can be routinely performed. The predominance of female and Caucasian patients limits generalisability, particularly to populations of colour who are at higher risk for hypertrophic or keloid scarring. This disparity underscores the need for larger and more demographically diverse cohorts in future validation studies.

Among assessed features, scar width emerged as the most robust predictor of 12‐month outcomes. Width showed both strong predictive signal (β = 0.80, p < 0.001) and high interrater reliability (ICC = 0.71), likely reflecting its relatively objective visual characteristics. These findings suggest that scar width predictably stabilises by 3 months post‐op and may serve as a dependable early indicator of long‐term scar spread.

Overall clinical opinion also demonstrated meaningful predictive capacity (β = 0.78, R ² = 0.31) with moderate reliability (ICC = 0.59). The binary classification of scars as ‘good’ or ‘poor’ was intentionally included to capture clinician gestalt, reflecting real‐world decision‐making that integrates multiple visual cues [18]. Although subjective, this measure demonstrated meaningful prognostic value, particularly in identifying scars unlikely to improve without intervention. The moderate interrater agreement highlights the limitations of subjective assessment alone and supports the complementary role of objective measurement tools.

Scar height showed moderate early predictability (β = 0.59, R ² = 0.19), reflecting partial stabilisation by 3 months in scars trending toward hypertrophy or flattening (Figure 9). Height, therefore, appears potentially useful for early risk stratification, although less definitively predictive than width. Pigmentation also showed high predictability with 87.5% matched pairs across time points (Figure 10). This is especially important in patients with skin of colour, in whom pigmentation changes may be more pronounced due to increased melanin activity and heightened susceptibility to postinflammatory dyspigmentation. However, limited representation in our cohort warrants cautious interpretation.

FIGURE 9.

Comparison of a scar at 3 months postoperative (left) and 12 months postoperative (right). The surgeon correctly predicted that the scar would flatten to a rating of 1 out of 6 for height.

FIGURE 10.

In between the 3‐month postoperative visit (left) and the 12‐month postoperative visit (right), the surgeon correctly predicted that pigmentation would decrease and the scar would fade despite its visibility early on.

In contrast, vascularity and depth demonstrated weaker associations and lower interrater reliability, consistent with the biologic variability of inflammatory and remodelling processes during early wound healing [19]. These findings suggest that early vascular prominence and perceived depth at 3 months remain biologically dynamic, and substantial improvement may occur. Low interrater agreement further suggests that these factors are challenging to assess consistently using visual examination alone. These findings caution against premature vascularity‐ and depth‐based intervention based solely on these parameters.

Objective technologies such as colorimetry or ultrasound‐based thickness measurement may improve reproducibility for less reliable features; however, such tools are not routinely available in standard postoperative clinics. These findings highlight the trade‐off between feasibility and objectivity in early scar assessment and support the development of integrated approaches.

Results from a separate analysis by our group demonstrated significant variation in pigmentation outcomes by skin type, with Fitzpatrick V‐VI scars exhibiting increased pigmentation (+0.35 ± 1.89), while lighter skin tones showed overall reduction [20]. Black and Hispanic patients were also less likely to achieve favorable pigmentation outcomes compared to Caucasian patients (p < 0.07). These findings support delaying pigment‐targeted interventions until full scar maturation, particularly in patients of colour who may be at higher risk of adverse effects from aggressive early therapies [21, 22, 23, 24].

These findings also carry implications for addressing disparities in scar outcomes across racial and ethnic groups. Individuals of African and Asian descent are known to be at increased risk for hypertrophic scarring and keloid formation [25]. Early identification of high‐risk scars using predictive features may enable proactive monitoring and intervention in susceptible populations, contributing to more equitable resource distribution and improved reconstructive outcomes.

Substantial interrater variability underscores the persistent challenge of subjectivity in clinical scar evaluation. Emerging image‐based analytic tools, including artificial intelligence, may offer complementary standardisation; however, such systems are trained on human‐labelled datasets and therefore inherently reflect observational variability. In a separate prospective cohort study, a Smart Image Analysis model applied to scar photographs demonstrated high accuracy in binary classification and strong predictive correlations for width and height, with more modest performance for vascularity [26]. These findings suggest that AI may serve as an adjunctive validation tool rather than a replacement for clinical assessment [27]. The present study establishes a necessary clinician‐based benchmark to inform future validation efforts in larger, more diverse cohorts.

Interrater reliability emerged as a critical determinant of feasibility. Features with higher reliability appear better suited for early predictive frameworks, whereas those with poor agreement may require delayed assessment or objective augmentation. These findings suggest that reliability may be a prerequisite for meaningful early prognostication. Future studies should prioritise larger, multicentre cohorts with greater demographic diversity, standardised scar management, and integration of objective or image‐based tools to advance early scar assessment from exploratory analysis to validated prediction models.

5. Conclusion

The appearance of a scar undergoes several phases of change and maturation as it heals. The ability to predict the final appearance of a scar at 12 months or longer postinjury would prove beneficial for adjusting study and treatment timelines. Our prospective single‐site study enrolled 40 participants who underwent various surgical procedures that resulted in scars. Although predictions regarding vascularity and depth were minimally accurate, other factors such as height, width, and pigmentation showed more promise in their predictive value.

Funding

We would like to acknowledge the Crown Family Philanthropies for their support of this project. Use of the SCAR‐Q was made under licence from McMaster University, Hamilton, Canada.

Ethics Statement

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.