From Assessment to Action: A Clinical Decision Support Application Translating SIPAT into Multidomain Psychosocial Risk Profiles in Lung Transplant Candidates

1. Background

Psychosocial assessment constitutes a core component of the evaluation process for solid organ transplantation, including lung transplantation. Beyond biomedical eligibility, successful transplantation depends on a range of behavioral and psychosocial factors, including treatment adherence, substance use, psychological stability, and the availability of social support. These factors have been consistently associated with post-transplant outcomes, including graft survival, complications, and quality of life [3,4,5]. In lung transplantation, psychosocial risk assessment is particularly relevant because candidates often present with advanced chronic respiratory disease, functional limitation, prior smoking exposure, psychological distress, and complex long-term treatment demands.

The Stanford Integrated Psychosocial Assessment for Transplantation (SIPAT) is one of the most widely used structured instruments for assessing psychosocial readiness in transplant candidates [1,2]. It provides a comprehensive, multidimensional evaluation across domains such as readiness for transplantation, social support, psychological functioning, and substance use. The use of SIPAT has improved standardization of psychosocial assessment and facilitated communication within multidisciplinary transplant teams.

Despite these advantages, SIPAT remains primarily a descriptive and score-based instrument. Its interpretation requires clinical expertise, and translating item-level scores into actionable clinical decisions may be challenging in routine practice. In particular, SIPAT does not directly provide individualized estimates of psychosocial risk or prioritize domains that may require targeted intervention. Consequently, its application may be limited to general candidate categorization rather than dynamic support of clinical decision-making and psychosocial prehabilitation.

At the same time, there is increasing recognition that psychosocial assessment should not serve solely as a gatekeeping mechanism but also as a foundation for targeted intervention [7]. The concept of psychosocial prehabilitation emphasizes the identification and modification of risk factors prior to transplantation, aiming to improve readiness and optimize outcomes [15,16]. However, implementing this approach in practice requires tools that not only assess risk but also structure and prioritize it in a clinically meaningful way.

Advances in digital health and clinical decision support systems offer a potential solution to this gap. By integrating structured assessment data with algorithmic processing and visualization, such systems can facilitate interpretation of complex multidimensional information and support individualized care planning [10,11,12]. In transplant medicine, this approach may enhance the practical utility of psychosocial assessment by translating scores into actionable risk profiles.

The present study describes the development and evaluation of a clinical decision support application designed to translate SIPAT-based psychosocial assessment into multidomain individualized risk profiles in lung transplant candidates. The application integrates SIPAT item-level data with probabilistic risk estimation, visualization, and cohort-referenced interpretation to support clinical decision-making and psychosocial prehabilitation. In addition, the study examines the internal structure of generated risk profiles, their associations with SIPAT domains, and their potential clinical interpretability.

2. Materials and Methods

2.1. Study Design and Data Source

This study was based on a retrospective dataset of adult lung transplant candidates evaluated at a single tertiary transplant center. Psychosocial assessments were conducted as part of routine clinical practice using the SIPAT instrument. The dataset included 496 lung transplant candidates with complete or near-complete data on SIPAT items and relevant variables required for application development.

All data were anonymized prior to analysis. The study was conducted in accordance with applicable ethical standards for retrospective analyses of clinical data.

2.2. Psychosocial Assessment

Psychosocial functioning was assessed using the SIPAT instrument, which evaluates multiple domains relevant to transplant candidacy. The instrument consists of item-level ratings grouped into four primary domains: readiness for transplantation, social support, psychological functioning, and substance use.

Each item is scored according to predefined criteria, reflecting the severity or presence of specific psychosocial factors. Domain scores are calculated by summing item-level values within each domain, and a total SIPAT score provides an overall index of psychosocial risk.

For analytical purposes, both item-level data and aggregated domain scores were used. Domain-level summaries included readiness, social support, psychological functioning, and substance use, as well as the total SIPAT score.

2.3. Development of the Clinical Decision Support Application

A clinical decision support application was developed using a Python-based framework with an interactive web interface. The application accepts patient age and SIPAT item scores as input variables and generates a structured psychosocial risk profile.

Risk estimation was performed using random forest–based models designed to transform SIPAT item-level data into structured probabilistic representations of psychosocial risk. To improve the interpretability of model outputs, probability calibration was applied, allowing model outputs to be interpreted as estimated risk values.

Predicted probabilities were expressed as percentage-based risk estimates and presented across multiple domains. These estimates should be interpreted as structured transformations of SIPAT-derived information rather than independent predictive outputs.

Non-adherence risk was modeled as a second-stage outcome, using previously estimated domain-specific risks as input variables. This approach allowed integration of multiple psychosocial dimensions into a composite indicator of adherence-related risk.

Cognitive functioning was represented using a scaled transformation of the SIPAT cognitive subcomponent. In addition to probabilistic outputs, the application incorporates rule-based elements, including SIPAT total score classification and domain-level burden scoring.

The application interface provides numerical outputs, categorical risk levels, and graphical visualizations, including bar charts and radar plots, facilitating rapid interpretation of psychosocial profiles.

2.4. Risk Categorization

For clinical interpretability, estimated risk values were categorized into three levels: low risk, defined as less than 20 percent; moderate risk, defined as 20 to 49 percent; and high risk, defined as 50 percent or greater. These thresholds were applied consistently across all domains to support standardized interpretation.

2.5. Domain-Level Burden and Cohort Referencing

Domain-level burden scores were calculated by summing item-level SIPAT scores within each domain. To enable contextual interpretation, domain scores were standardized relative to cohort-level reference values using z-scores.

Z-scores were computed using the mean and standard deviation of each domain within the study population, allowing identification of domains with relatively elevated or reduced burden compared to the reference cohort. This approach supports prioritization of psychosocial interventions at the individual level.

2.6. Statistical Analysis

Descriptive statistics were calculated for all estimated risk domains, including mean values, standard deviations, and distribution across predefined risk categories.

Associations between estimated risk values and SIPAT domain scores were assessed using Spearman rank correlation coefficients, given the non-normal distribution of variables.

To explore the internal structure of psychosocial risk profiles, unsupervised clustering was performed using k-means clustering applied to standardized risk variables. The number of clusters was selected based on interpretability and stability of solutions. Identified clusters were characterized based on their dominant risk features.

Feature importance was assessed for each model using impurity-based importance measures derived from the underlying random forest classifiers. For calibrated models, importance values were averaged across component estimators.

All analyses were conducted using Python and standard data science libraries.

3. Results

3.1. Study Sample

The analysis included 496 lung transplant candidates with available SIPAT-based psychosocial assessments. Overall psychosocial burden in the cohort was low, with most patients presenting minimal scores across individual SIPAT domains. The distribution of SIPAT-derived variables suggested that clinically relevant psychosocial risk was concentrated within relatively small subgroups rather than uniformly distributed across the population.

3.2. Distribution of Estimated Psychosocial Risk

Estimated risk values were strongly skewed toward low-risk categories across all domains. The majority of patients were classified as low risk, while moderate-risk categories were rare and high-risk groups were relatively small but clinically distinct.

High risk for depression was identified in 6.5% of patients, for anxiety in 2.2%, for nicotine-related risk in 11.3%, for alcohol-related risk in 4.4%, for illicit drug use in 2.2%, and for social support deficits in 8.1%. High non-adherence risk was identified in 1.4% of the cohort.

Moderate-risk classifications were infrequent, resulting in a predominantly bimodal distribution of risk, with patients clustering in either low- or high-risk categories. This pattern was most pronounced for anxiety and substance-related domains, where intermediate levels were rarely observed.

Descriptive statistics of continuous risk estimates further supported this distribution, with low median values across all domains and large standard deviations reflecting the presence of a limited number of high-risk cases.

3.3. Multidomain Psychosocial Risk Profiles

Unsupervised clustering of multidomain risk estimates identified three distinct psychosocial profiles within the cohort.

The largest cluster (n = 432) represented a low-risk profile, characterized by consistently low estimated probabilities across all domains. This group accounted for approximately 87% of the cohort and reflected patients with minimal psychosocial burden.

A second cluster (n = 53) was defined by markedly elevated nicotine-related risk (mean approximately 96%), accompanied by moderate elevations in other domains. This profile suggests a subgroup in which substance-related factors, particularly tobacco use, represent the dominant psychosocial concern.

The smallest cluster (n = 11) demonstrated a high psychosocial burden across multiple domains. Patients in this group exhibited substantially elevated risks for depression (approximately 90%), anxiety (approximately 82%), and reduced social support (approximately 67%), along with increased non-adherence risk (approximately 28%). This cluster represents a clinically vulnerable subgroup with complex and multidimensional psychosocial needs.

Together, these findings indicate that psychosocial risk is structured into distinct and clinically interpretable profiles rather than distributed along a simple linear gradient.

3.4. Associations with SIPAT Domains

Estimated risk values showed statistically significant correlations with SIPAT domain scores, supporting the internal coherence of the application.

Depression and anxiety risks demonstrated strong associations with the psychological domain and total SIPAT score, with Spearman correlation coefficients reaching up to 0.80. Social support risk was strongly associated with the social support domain (rho = 0.77), indicating domain-specific alignment between input assessment and generated risk estimates.

Substance-related risks, including nicotine, alcohol, and drug use, showed moderate to strong associations with the substance use domain and total SIPAT score. These associations suggest that the application preserves the multidimensional structure of SIPAT while translating it into probabilistic outputs.

Non-adherence risk demonstrated moderate correlations with all SIPAT domains, with the strongest association observed for the total SIPAT score (rho = 0.48). This pattern is consistent with the integrative nature of adherence-related risk, which reflects multiple aspects of psychosocial functioning.

All reported correlations were statistically significant (p < 0.001).

3.5. Feature Importance

Feature importance analysis demonstrated that risk estimation was primarily driven by domain-specific SIPAT items, with patterns consistent across models.

Depression and anxiety risks were most strongly influenced by items related to psychopathology, including depression and anxiety severity components. Nicotine-related risk was predominantly driven by the SIPAT nicotine use item, while alcohol-related risk was primarily associated with alcohol use and relapse risk items.

Drug-related risk was driven by substance use and relapse indicators, and social support risk was most strongly influenced by items assessing availability and functionality of support as well as living environment.

These findings indicate that the models rely on clinically meaningful input variables and preserve the conceptual structure of the SIPAT instrument rather than generating arbitrary associations.

For non-adherence risk, the most influential predictors were previously estimated risks for anxiety, drug use, and social support deficits. This supports the interpretation of non-adherence as a multidimensional construct influenced by both psychological and social factors.

3.6. Application Output and Clinical Utility

The application generated a structured and interpretable output for each patient, integrating SIPAT scoring, multidomain risk estimation, and graphical visualization.

Outputs included total SIPAT score, candidate category, domain-specific risk estimates, and visual representations of psychosocial burden. Risk estimates were presented as both continuous values and categorical levels, facilitating rapid clinical interpretation.

Domain-level burden was contextualized using cohort-referenced z-scores, allowing identification of domains with relatively elevated psychosocial load compared to the study population. In addition, the application provided a ranked summary of intervention priorities based on domain-level burden, supporting targeted psychosocial prehabilitation planning.

Overall, the application translated complex psychosocial assessment data into a structured format that enables identification of high-risk subgroups and prioritization of clinically relevant interventions.

4. Discussion

The present study describes the development and evaluation of a clinical decision support application designed to translate SIPAT-based psychosocial assessment into structured, multidomain risk profiles. The findings demonstrate that the application provides a coherent and clinically interpretable representation of psychosocial risk, preserving the multidimensional structure of the underlying assessment while enhancing its practical utility.

A key observation is that psychosocial risk in the studied cohort was not uniformly distributed but instead concentrated within relatively small, clinically distinct subgroups. The majority of patients were classified as low risk across domains, while high-risk profiles were limited to a minority of cases. This pattern is consistent with clinical experience in transplant populations and prior literature indicating that psychosocial risk factors tend to cluster within specific vulnerable subgroups [3,4].

Importantly, the application identified distinct multidomain risk profiles, including a subgroup characterized by elevated substance-related risk and a smaller cluster with high overall psychosocial burden. These profiles reflect clinically meaningful patterns that are not immediately apparent from raw SIPAT scores alone. In particular, the identification of a high-burden subgroup with combined psychological distress, reduced social support, and elevated non-adherence risk highlights the potential value of structured risk profiling in guiding targeted interventions [8,9].

The observed associations between estimated risk values and SIPAT domains support the internal consistency of the application. Risk estimates were strongly aligned with their corresponding domains, indicating that the transformation of SIPAT data into probabilistic outputs preserves the conceptual framework of the original instrument [1,2]. This is further supported by feature importance analyses, which demonstrated that domain-specific SIPAT items were the primary drivers of risk estimation. Together, these findings suggest that the application does not introduce arbitrary relationships but rather formalizes and operationalizes clinically meaningful information already embedded in the assessment.

From a clinical perspective, the primary contribution of the application lies in its ability to shift psychosocial assessment from a descriptive process toward a structured, actionable framework. Traditional use of SIPAT often focuses on overall categorization of transplant candidates, which may implicitly function as a gatekeeping mechanism. In contrast, the approach presented here emphasizes identification of modifiable risk factors and prioritization of psychosocial prehabilitation [7,15,16]. By providing individualized risk estimates, visual summaries, and domain-based prioritization, the application supports more targeted and proactive clinical decision-making.

The integration of cohort-referenced domain scores represents an additional strength. By contextualizing individual patient results relative to a reference population, the application facilitates interpretation of psychosocial burden beyond absolute scores. This approach is consistent with broader trends in clinical decision support, where contextualized and interpretable outputs are essential for effective implementation in practice [10,11,12].

It is important to note that the probabilistic outputs generated by the application should not be interpreted as independent predictive models in the conventional sense. Given the structured relationship between SIPAT items and defined outcomes, the model outputs are best understood as a transformation of assessment data into a probabilistic format that enhances interpretability. This distinction is critical to avoid overinterpretation of model performance and aligns the application with the principles of clinical decision support rather than predictive modeling [13,14].

Several limitations should be considered. First, the study was conducted in a single-center cohort, which may limit generalizability to other transplant populations. Second, the retrospective design precludes assessment of prospective clinical impact, including whether use of the application improves decision-making or patient outcomes. Third, although the application incorporates machine learning methods, it does not provide independent validation against external clinical endpoints. Furthermore, because the application was developed using data from lung transplant candidates, its generalizability to other solid organ transplant populations requires further evaluation. Future studies should address these limitations through multicenter validation and longitudinal analyses assessing the impact of risk-guided interventions.

Despite these limitations, the proposed approach offers a practical framework for enhancing the clinical utility of psychosocial assessment in lung transplantation. By translating SIPAT data into structured risk profiles and intervention priorities, the application supports a shift toward more individualized and proactive psychosocial care.