For the purposes of analysis, we categorized patients with a loneliness score higher than the mean as “lonely.” We called subjects with a loneliness score lower than or equal to the mean “nonlonely.”
Total Hospital Visits. We determined the total number of hospital visits by using a combination of patient self-report and medical record data analysis. We obtained self-report information from the questionnaire. The computerized medical records listed the number of ED visits, admissions after an ED visit, and direct admissions to the hospital without being processed through the ED. We determined total hospital visits by adding the number of ED visits and direct admissions. When patients indicated that they used more than one hospital, we used the higher number of self-reported visits or medical-record-determined visits as the total hospital visit score. If a patient indicated that they only use the Lawrence General Hospital, we used that medical record to determine total hospital visits. We collected data in a blinded fashion, using medical record numbers only to indicate patient identity.
Reason for Visit. We performed a blind review of medical records to obtain the chief complaint and discharge diagnosis from the ED. We categorized the patients according to the organ system involved (eg, cardiac, obstetric, pulmonary, and so forth).
Chronic Illness. We reviewed the medical records for past medical history. We noted any illness that might contribute to increased ED use identified it by organ system.
Statistical Analysis
We examined loneliness score, total visits, and hospital admission with least mean square linear regressions to evaluate for a relationship among loneliness, total hospital visits, and hospital admissions. We compared these characteristics of lonely and nonlonely subjects using a 2-tailed t test for significance. We used chi-square analysis to compare patient characteristics in the lonely and nonlonely groups (P •.05 was considered significant).
Baseline Characteristics of Patients
One hundred eighty-two patients were enrolled in our study; 164 participated. Eighteen were excluded because of language other than English or Spanish, dementia, delirium, psychosis, age younger than 14 years, or failure to answer at least 17 of the 20 UCLA survey questions. Twenty additional people refused enrollment.
Of the 164 patients studied, 42% were men. The mean loneliness score was 39.06 with a standard deviation of 12, which is the same as in a normal population. There was no statistical association between the sex of a patient or presence of a primary physician and being lonely (P >.20). Patients who were lonely did not have increased underlying chronic illness (P = .56) or major differences in their reasons for the visit. Twenty-five percent of the patients spoke Spanish. There was a statistical significance between having a preference for the Spanish language and increased loneliness score (P = .001).
Hospital Use
Figure 2 shows the statistically significant association between loneliness score and total hospital visits (b coefficient = 0.063; standard error = 0. 01475; correlation coefficient = 0.32; P <.001; 95% confidence interval, 0.034 - 0.093). There were 74 patients with loneliness scores above the mean of 39 who used the hospital an average of 3.6 times a year. The remaining 90 patients with loneliness scores lower than or equal the mean used the hospital an average of 2.2 times a year (P = .009). There was no significant association between loneliness score and number of hospital admissions (P = .52).
Discussion
We found a significant association between a patient’s loneliness score and total hospital visits. We do note, however, that there were 5 outlying data points representing subjects who used the ED more than 9 times in a year. These can have a large effect on the regression output. All of these patients fell into the category of “lonely” by our definition, which seems significant. To further explore the effect of these outliers on our results, we excluded them and performed another linear regression on our data Figure 3. We found that though our b coefficient was lower (0.037), it was still statistically significant (P = .003).
Two surprising results emerge from our data. First, we had expected that there would be more chronic illness in those who were lonely, given the recent studies associating loneliness with increased illness. Second, we expected that our ED population would be lonelier than the population at large. Neither of these expectations proved true. Thus, it seems that although lonely people use the ED more often, they are not necessarily more ill. The second finding is more difficult to explain. It is possible that our study sample contained a population of people who were less lonely but more ill. Elderly people score lower than younger people when they are given the loneliness questionnaire.5 The older population (aged >60 years) in our study had lower loneliness scores, though this was not statistically significant (P = .12). Elderly people also tend to be more ill.8 We might interpret our data to indicate that though an ED has an increased number of lonely patients, it also has an increased number of less lonely elderly patients. The net result is a prevalence of loneliness in the ED that equals that of a normal population.