Background In patients with coronary heart disease (CHD), loneliness is associated with increased risk of morbidity and mortality. No predictive tool is available to detect patients who are influenced by loneliness to a degree that impacts mortality. Aim To: (i) propose a prediction model that detects patients influenced by loneliness to a degree that increases one-year all-cause mortality, (ii) evaluate model classification performance of the prediction model, and (iii) investigate potential questionnaire response errors. Method A cohort of patients with CHD (n = 7169) responded to a national cross-sectional survey, including two questions on loneliness. Information on cohabitation and follow-up information on one-year all-cause mortality were obtained from national registers. Prediction model development was based on the prognostic values of item responses in the questionnaire on loneliness and of cohabitation, evaluated with Cox-proportional Hazards Ratio (HR). Item responses which significantly predicted one-year mortality were included in the high-risk loneliness (HiRL) prediction model. Sensitivity, specificity and likelihood ratio were calculated to evaluate model classification performance. Sources of response errors were evaluated using verbal probing technique in an additional cohort (n = 7). The TRIPOD checklist has been used to ensure transparent reporting. Results Two item responses significantly predicted one-year mortality HR = 2.24 (95%CI = 1.24-4.03) and HR = 2.65 (95%CI = 1.32-5.32) and were thus included in the model. Model classification performance showed a likelihood ratio of 1.89. Response error was evaluated as low. Conclusion Based on the prognostic value in a loneliness questionnaire, a prediction model suitable to screen patients with CHD for high-risk loneliness was suggested. Relevance to clinical practice The HiRL prediction model is a short and easy-to-use screening tool that offers clinical staff to identify patients with CHD who are influenced by loneliness to a degree that impacts mortality. However, further evaluation of model performance and questionnaire validation is recommended before integrating the model into clinical practice.