Social disconnection is usually measured after it has become obvious. This study takes a different approach: use thirteen questions to detect threats to connection earlier, before isolation and loneliness fully take hold. Social interactions are essential to social connectedness among older adults, but many existing scales are narrow in scope and may not be encompassing, practical, or relevant across clinical and community settings.

The need, then, is for more sensitive scales that identify “upstream risk,” potentially allowing timely referral or intervention. Social connectedness includes structural aspects such as network size, functional aspects such as perceived social support and loneliness, and relationship quality or strain.

It is an umbrella term covering social isolation, which means objectively limited contact, and loneliness, which means the subjective feeling of being alone. An estimated twenty-five percent of older adults are socially isolated, and more than forty percent are lonely; these concepts are related but do not necessarily overlap completely.

Because isolation and loneliness are linked with diminished physical and mental health, cognitive impairment, and risky behaviors, the study argues for early identification through routine screening. Existing scales have helped define and quantify risk, but many may not be encompassing, practical, or relevant across clinical and community settings.

Used independently, these scales may identify separate dimensions without capturing the complexity of existing risk or guiding intervention opportunities. The proposed solution is a more encompassing and sensitive scale that identifies “upstream risk,” including emerging threats to social connection.

Here, social interactions are divided into physical opportunities to interact and the emotional fulfillment resulting from those interactions. A sequential process first developed an initial instrument and then methodologically reduced it to a brief thirteen-item scale.

The process gathered existing validated scales and items, then used statistical analyses to find the most parsimonious set measuring threats to social connectedness. The goal of a parsimonious brief measure was to increase practical administration in research and practice settings.

The scales were critically reviewed for content and overlapping concepts to optimize relevant items while reducing redundancies. Three research and practice experts ranked each item for relevance to social connectedness. When consensus was not reached, the experts retained the item to remain more inclusive at this stage.

The study used a cross-sectional, internet-delivered questionnaire for adults aged sixty years and older, recruited nationwide through a Qualtrics Internet Panel from June twenty nineteen to September twenty nineteen. Quota sampling parameters were used to support diversity in age, sex, race, and geography despite the potential sampling bias of convenience sampling.

Of four thousand one hundred one completed surveys, nineteen were omitted for missing data, leaving four thousand eighty-two participants from all fifty states and two U.S. territories. The data were analyzed in R using several methods to assess dimensionality and construct risk factors involving limited opportunity for social interactions and lacking emotional fulfillment.

First, a unidimensional item response theory model assessed each item’s relationship with risk and generated participant theta scores; IRT then reduced the initial thirty-six items to thirteen. Clustering techniques assessed subgroups based on response patterns, and structural equation modeling confirmed relationships among the latent variables measured by the thirteen-item scale.

Table 2 describes the four thousand eighty-two older adults included in the study. The average age was sixty-nine point fifty-eight, with an average of three point twenty-nine chronic conditions; the table also reports sex, ethnicity, race, education, and whether participants lived with a spouse or partner.

These characteristics matter because they define the population used to develop and evaluate the social-interaction risk screener and its latent-factor and clustering analyses. Overall model fit was better for the thirteen-item scale than for the initial thirty-six items.

The two-parameter logistic model was selected because its discrimination values supported selecting items relevant across the trait continuum. The one-parameter model was tested for parsimony, while the three-parameter model addressed random guessing, but the added parameterization did not outweigh its weaker statistical and conceptual fit.

The lack of a sizable increase from the guessing parameter made the two-parameter logistic model the most appropriate model for these data. Table Three compares one-, two-, and three-parameter IRT models for the initial thirty-six items and the thirteen-item scale, reporting log likelihood, parameter counts, AIC, and BIC.

For the thirteen-item scale, the two-parameter model has an AIC of forty thousand eight hundred nineteen and a BIC of forty thousand nine hundred eighty-four, while the three-parameter model has an AIC of forty thousand eight hundred thirty-seven and a BIC of forty-one thousand eighty-three.

The authors selected the two-parameter model because item discrimination supported coverage across the trait continuum, while the three-parameter model tested random guessing. When binary responses from the thirteen-item scale were summed into a total score ranging from zero to thirteen and correlated with the IRT theta parameter, the correlation was strong and significant: r equals 0.896, with p less than 0.001.

This result indicates that the summed binary-response score can serve as a statistical proxy for the older adult’s IRT score, supporting more practical use in research and practice. Three measures evaluated the number of clusters based on older adults’ responses to the thirteen items, comparing clustering methodologies through connectivity, similarity, and compactness.

Connectivity and silhouette width should be minimized, while the Dunn index should be maximized. Taken together, the results indicated that the optimal number of clusters was two, although the Dunn index has a bias toward a larger number of clusters.

Figure one plots component scores for the two-cluster solution after scaling them into two dimensions. The colored point clouds overlap substantially, but the clusters show clearer separation along the first dimension, while separation along the vertical second dimension is more subtle.

This matters because, as the authors note, cluster separation is a key consideration when using the model to classify individuals in an applied setting. The structural equation model fit exceptionally well, with an RMSEA of 0.048 and a CFI of 0.954.

Within the structural regression component, the contribution of the Factor 3 items exceeded the contribution of the Factor 2 items, with coefficients of 1.348 and 0.832. Figure 3 presents a structural equation model linking three factors related to social connectedness risk.

Factor 1 represents general feelings of disconnection through items such as “I feel isolated from others,” while Factors 2 and 3 capture physical opportunity and emotional fulfillment, respectively; the diagram also shows covariance between Factors 2 and 3. This matters because the model identifies the general-disconnectedness items as prediction targets and shows how upstream physical and emotional conditions relate to that risk.

The SEM confirmed three distinct U-SIRS-13 factors: general feelings of disconnectedness, physical opportunities for social interaction, and emotional fulfillment from social interactions, or the lack of it. The physical opportunity and emotional fulfillment sub-scales roughly predicted each other, with r equal to 0.59, and both strongly predicted general feelings of social disconnectedness.

The model therefore distinguishes having an opportunity to interact from perceiving emotional fulfillment from that interaction. Despite three distinct factors, IRT showed that all thirteen items conform to a singular trait, supported by a Cronbach’s alpha of 0.80 for binary-scored data.

The recommended scoring is a continuous count variable of risk rather than independent sub-scale scores, while item-level risk can help practitioners identify threats to structure, function, or quality aspects of social connection. The IRT theta score strongly correlated with the number of items endorsed as risk: r equals 0.896, with p less than 0.001, across a binary-scored count ranging from zero to thirteen.

That result validates using the total count score in practice, with higher values indicating more risk after the thirteen items are summed. Table 6 presents the recommended items, response choices, and practical scoring for dichotomizing responses to identify the maximum amount of upstream risk.

Across demonstration studies and evaluation efforts, practical scoring showed consistently strong internal reliability, with Cronbach’s alpha ranging from 0.78 to 0.85. Table six gives a practical scoring key for the thirteen U-SIRS items. Each response—“Never,” “Sometimes,” or “Often”—is assigned either zero or one, with the item scores summed into a count from zero to thirteen; higher values indicate more risk.

This approach is supported by the reported correlation of point eight nine six between the IRT theta score and the binary item count, suggesting that the table translates the model into an easy-to-use measure. Table seven reports internal reliability for the practically scored U-SIRS-thirteen across nine samples, including the current nationwide survey and several intervention and evaluation studies.

Sample sizes range from forty-four to four thousand eighty-two, while Cronbach’s alpha ranges from zero point seventy-eight to zero point eighty-five. The consistency of these values across varied older-adult populations supports continued use of the scale, while also motivating replication in more diverse samples.

Although the study included four thousand eighty-two diverse older adults aged sixty years and older across the United States, probabilistic sampling was not used, so the data were not nationally representative. Internet-based recruitment and data collection may have introduced selection bias related to technology access, education level, and affluence, potentially excluding people at greater risk for social disconnectedness.

The analytic sample may therefore not generalize to the overall older adult population, especially people with lower socioeconomic status, and self-reported data may be subject to social desirability bias. Strengths included diverse professional input for item selection, robust statistical analyses, and emerging evidence that the practically scored scale can be replicated.

The U-SIRS-13 is a thirteen-item scale created from items in seven previously validated scales to document threats to social connection. Its three interrelated sub-scales measure general disconnectedness, physical opportunities for social interactions, and emotional fulfillment from social interactions or the lack of it.

Although the sub-scales are distinct, IRT and SEM support using the instrument as a single scale, and the correlation between theta scores and summed binary-scored items supports practical utilization. Future efforts still need to identify risk levels and thresholds that classify risk and guide referrals to appropriate programs and services.

The U-SIRS-13 combines three related dimensions of social connection into one practical thirteen-item risk count. Its evidence supports screening for upstream threats, while future work still needs to establish risk thresholds.