Study staff will provide a combination of 8 social interaction visits and phone calls to the participants randomized to this condition.

The RAND 36-Item Health Survey is a set of generic, coherent, and easily administered quality-of-life measures. Each item is scored on a 0 to 100 range. Higher scores indicate better functioning. The social functioning subscale was calculated using the average of 2 of the 36 items (item 20, 32)

The RAND 36-Item Health Survey is a set of generic, coherent, and easily administered quality-of-life measures. Each item is scored on a 0 to 100 range. Higher scores indicate better functioning. The emotional functioning subscale was calculated using the average of 3 of the 36 items (item 17, 18, 19)

Health service use is measured with the Cornell Service Index, a questionnaire that includes both clinical and informal community-based health services and has been used in depression care studies among older adults. The index asks whether an individual has used a list of services in the 4 months prior, what type of provider was seen, the site of the service, and reasons for service use. Numbers reported indicate the number of participants who reported hospital and emergency service use at specified timepoints.

Engagement (adherence) was measured by percentage of attendance of meetings as reported by the peer mentors, with a threshold of 80% of meetings attended.

Working alliance was measured on a weekly basis only in the intervention arm with the Working Alliance Inventory-SF. The survey has 12 items with subscales that measure affective bond, and agreement on tasks and goals and are scored using a Likert scale. The goal subscore was calculated by summing items 1, 6, 8, and 11. The subscale ranges from scores of 4-28, with higher scores indicating better agreement on goals.

Working alliance was measured on a weekly basis only in the intervention arm with the Working Alliance Inventory-SF. The survey has 12 items with subscales that measure affective bond, and agreement on tasks and goals and are scored using a Likert scale. The bond subscore was calculated by summing items 3, 5, 7, and 9. The subscale ranges from 4-28. A higher subscore indicates higher affective bond.

Working alliance was measured on a weekly basis only in the intervention arm with the Working Alliance Inventory-SF. The survey has 12 items with subscales that measure affective bond, and agreement on tasks and goals and are scored using a Likert scale. The task subscore was calculated by summing items 2, 4, 10, and 12. The subscale ranges from 4-28. A higher subscore indicates higher agreement on tasks.

The 20-item R-UCLA loneliness scale measures one's subjective feelings of loneliness as well as feelings of social isolation. Items are summed to create a score. The scores range from 20-80 and higher scores indicating higher level of loneliness. A total score of 20-34 represents low degree of loneliness; 35-49 represent moderate degree of loneliness; 50-64 represent moderately high degree of loneliness; and 65 or above represent high degree of loneliness.

The General Self-Efficacy Scale assess a general sense of perceived self-efficacy with the aim in mind to predict coping with daily hassles as well as adaptation after experiencing all kinds of stressful life events. GSES score is calculated by summing all item scores. The score ranges from 10-40 and higher scores indicating more self-efficacy.

The Brief-COPE is a 28 item self-report questionnaire designed to measure effective and ineffective ways to cope with a stressful life event. Reliance on adaptive coping strategies were calculated as the sum of the following subscales: active coping (item 2, 7), emotional support (item 5, 15), use of informational support (item 10, 23), positive reframing (item 12, 17), planning (item 14,25), acceptance (item 20, 24), religion (item 22, 27). Each of the subscales ranges from 2-8. The range of scores for adaptive coping is 14-56, with higher scores indicating higher adaptive coping.

A study staff member will provide eight weekly social interaction phone calls to the depressed older adult.

Social interaction: Study staff will provide a social interaction phone calls to the participants randomized to this condition.

Total of all reporting groups

Depression will be assessed with the Patient Health Questionnaire (PHQ)-9. The PHQ-9 score is summed. There are 9 items in the questionnaire and scores range from 0 to 27. A score of 5 indicates mild depressive symptoms, 10 - 14 indicates moderate depression, 15-19 indicates moderately severe depression, 20-27 indicates severe depression.

A study staff member will provide eight weekly social interaction visits and phone calls to the depressed older adult.

Social interaction: Study staff will provide a combination of 8 social interaction visits and phone calls to the participants randomized to this condition.

The RAND 36-Item Health Survey is a set of generic, coherent, and easily administered quality-of-life measures. Each item is scored on a 0 to 100 range. Higher scores indicate better functioning.The physical functioning subscale was calculated using the average of 10 of the 36 items (item 3, 4, 5, 6, 7, 8, 9, 10, 11, 12)

All participants with at least one assessment completed

The RAND 36-Item Health Survey is a set of generic, coherent, and easily administered quality-of-life measures. Each item is scored on a 0 to 100 range. Higher scores indicate better functioning. The social functioning subscale was calculated using the average of 2 of the 36 items (item 20, 32)

All participants with at least one assessment completed

The RAND 36-Item Health Survey is a set of generic, coherent, and easily administered quality-of-life measures. Each item is scored on a 0 to 100 range. Higher scores indicate better functioning. The emotional functioning subscale was calculated using the average of 3 of the 36 items (item 17, 18, 19)

All participants with at least one assessment completed

Health service use is measured with the Cornell Service Index, a questionnaire that includes both clinical and informal community-based health services and has been used in depression care studies among older adults. The index asks whether an individual has used a list of services in the 4 months prior, what type of provider was seen, the site of the service, and reasons for service use. Numbers reported indicate the number of participants who reported hospital and emergency service use at specified timepoints.

All participants with at least one assessment completed

Engagement (adherence) was measured by percentage of attendance of meetings as reported by the peer mentors, with a threshold of 80% of meetings attended.

All participants with at least one assessment completed

Working alliance was measured on a weekly basis only in the intervention arm with the Working Alliance Inventory-SF. The survey has 12 items with subscales that measure affective bond, and agreement on tasks and goals and are scored using a Likert scale. The goal subscore was calculated by summing items 1, 6, 8, and 11. The subscale ranges from scores of 4-28, with higher scores indicating better agreement on goals.

All participants in the PEERS intervention with at least one assessment completed

Working alliance was measured on a weekly basis only in the intervention arm with the Working Alliance Inventory-SF. The survey has 12 items with subscales that measure affective bond, and agreement on tasks and goals and are scored using a Likert scale. The bond subscore was calculated by summing items 3, 5, 7, and 9. The subscale ranges from 4-28. A higher subscore indicates higher affective bond.

All participants in the PEERS intervention with at least one assessment completed

Working alliance was measured on a weekly basis only in the intervention arm with the Working Alliance Inventory-SF. The survey has 12 items with subscales that measure affective bond, and agreement on tasks and goals and are scored using a Likert scale. The task subscore was calculated by summing items 2, 4, 10, and 12. The subscale ranges from 4-28. A higher subscore indicates higher agreement on tasks.

All participants in the PEERS intervention with at least one assessment completed

The 20-item R-UCLA loneliness scale measures one's subjective feelings of loneliness as well as feelings of social isolation. Items are summed to create a score. The scores range from 20-80 and higher scores indicating higher level of loneliness. A total score of 20-34 represents low degree of loneliness; 35-49 represent moderate degree of loneliness; 50-64 represent moderately high degree of loneliness; and 65 or above represent high degree of loneliness.

All participants with at least one assessment completed

The General Self-Efficacy Scale assess a general sense of perceived self-efficacy with the aim in mind to predict coping with daily hassles as well as adaptation after experiencing all kinds of stressful life events. GSES score is calculated by summing all item scores. The score ranges from 10-40 and higher scores indicating more self-efficacy.

All participants with at least one assessment completed

The Brief-COPE is a 28 item self-report questionnaire designed to measure effective and ineffective ways to cope with a stressful life event. Reliance on adaptive coping strategies were calculated as the sum of the following subscales: active coping (item 2, 7), emotional support (item 5, 15), use of informational support (item 10, 23), positive reframing (item 12, 17), planning (item 14,25), acceptance (item 20, 24), religion (item 22, 27). Each of the subscales ranges from 2-8. The range of scores for adaptive coping is 14-56, with higher scores indicating higher adaptive coping.

All participants with at least one assessment completed

A study staff member will provide eight weekly social interaction phone calls to the depressed older adult.

Social interaction: Study staff will provide a social interaction phone calls to the participants randomized to this condition.

Null hypothesis: there is no difference in the mean SF36-physical function scores from baseline to 3-month for all study participants.

Degrees of freedom: 364

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-physical function scores from baseline to 6-month for all study participants.

Degrees of freedom: 364

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-physical function scores from baseline to 9-month for all study participants.

Degrees of freedom: 364

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-physical function scores from baseline to 12-month for all study participants.

Degrees of freedom: 364

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: There is no difference in the change of physical functioning over time between the control and PEERS participants.

Null hypothesis: there is no difference in the mean SF36-social function scores from baseline to 3-month for all study participants.

Degrees of freedom: 360

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-social function scores from baseline to 6-month for all study participants.

Degrees of freedom: 360

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-social function scores from baseline to 9-month for all study participants.

degrees of freedom: 360

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-social function scores from baseline to 12-month for all study participants.

Degrees of freedom: 360

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: There is no difference in the change of social functioning over time between the control and PEERS participants.

Null hypothesis: there is no difference in the mean SF36-emotional functioning scores from baseline to 3-month for all study participants.

Degrees of freedom: 366

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-emotional functioning scores from baseline to 6-month for all study participants.

Degrees of freedom: 366

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-emotional functioning scores from baseline to 9-month for all study participants.

Degrees of freedom: 366

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: there is no difference in the mean SF36-emotional functioning scores from baseline to 12-month for all study participants.

Degrees of freedom: 366

The a priori threshold for statistical significance is p=0.05.

A series of generalized linear mixed-effect models were implemented to predict the outcome measurements. The primary fixed effect predictors were group (control vs. PEERS), time (categorical), and group * time. Random terms in all models were: study participants and their interaction with time. The covariance of dependent variable values across time was modeled as a first-order auto-regression decaying function (AR1). Higher order fixed and random terms were pretested and removed from the model if nonsignificant.

Null hypothesis: There is no difference in the change of emotional functioning over time between the control and PEERS participants.