Dissecting an Online Intervention for Cancer Survivors

Abstract

The Internet has been used extensively to offer health education content and also for social support. More recently, we have seen the advent of Internet-based health education interventions that combine content with structured social networking. In many ways this is the Internet equivalent to small group interventions. While we have some knowledge about the efficacy of these interventions, few studies have examined how participants engage with programs and how that might affect outcomes. This study seeks to explore (a) the content of posts and (b) the nature of participant engagement with an online, 6-week workshop for cancer survivors and how such engagement may affect health outcomes. Using methodologies related to computational linguistics (latent Dirichlet allocation) and more standard statistical approaches, we identified (a) discussion board themes; (b) the relationship between reading and posting messages and outcomes; (c) how making, completing, or not completing action plans is related to outcome; and (d) how self-tailoring relates to outcomes. When considering all posts, emotional support is a key theme. However, different sets of themes are expressed in the first workshop post where participants are asked to express their primary concern. Writing posts was related to improved outcomes, but reading posts was less important. Completing, but not merely making, action plans and self-tailoring are statistically associated with future positive health outcomes. The findings from these exploratory studies can be considered when shaping future electronically mediated social networking interventions. In addition, the methods used here can be used in analyzing other large electronically mediated social-networking interventions.

Keywords

cancer survivorship engagement online interventions social support

In this study, four questions are examined using four sets of exploratory analysis (described below). The purpose of the first three analyses is to describe and link two key components of an online health education intervention with behavioral and or health status outcomes, such as changes in exercise, depression, illness intrusiveness, stress, and general health. Those two components consist of (a) themes discussed by participants and (b) participant engagement. The fourth analysis examines the effect of self-tailoring on health outcomes. These are secondary analyses using data from a recently published intervention study for cancer survivors (Bantum et al., 2014).

There have been many studies that have linked self-management programs to improvements in health behaviors. Some studies have identified the themes discussed by participants in both small-group and online behavioral interventions. However, these themes have seldom been linked to outcomes. There have also been a series of studies describing online engagement without linking engagement with outcomes. As we gain knowledge of these linkages, or lack thereof, we will be able to build more effective outcomes. In the following paragraphs, we review each of these components before describing how they may be linked.

Theme Analysis

In one study of 300 messages posted during an 8-week online support group for cancer survivors, the most common themes were exchange of information with other people who had survived cancer, discussion of symptoms, and frustration with health care providers (Klemm, 2008). Other work has focused on understanding gender differences in what is posted in online support groups for breast versus prostate cancer. Informational and emotional support accounted for nearly all the posts (Gooden & Winefield, 2007; Mo, Malik, & Coulson, 2009). “Facts about disease” was found as the most frequent theme in these groups. However, it is difficult to compare the themes between studies because the criteria for determining themes are not consistent.

Outcome Studies/Studies Linking Themes With Outcomes

There have been several studies linking participation in online cancer social support intervention with behavioral and or health status outcomes including participation in health care, depression, and stress (Gustafson et al., 2001; Winzelberg et al., 2003; Wise, Han, Shaw, McTavish, & Gustafson, 2008). Some of these authors and others have examined the links between what is being said during these interactions and outcomes. One of the mechanisms that is thought to be related to better outcomes for cancer survivors is emotional expression (Berry & Pennebaker, 1993). A series of studies by Shaw et al. (2006) and later Lieberman and Winzelberg (2009) examined the role of religious expression on emotional outcome for online cancer survivor groups. While Shaw et al. found an association, Lieberman and Winzelberg did not. In another study, Lieberman and Goldstein (2006) found that the expression of different negative emotions, that is, fear versus anger, were associated with different outcomes. Lieberman (2007) also found that the process of trying to make sense out of a traumatic experience (what he termed insightful disclosure) led to breast cancer survivors having less breast-cancer worry and increased functional well-being.

Other Engagement Studies

Research has focused on reporting engagement as number of times logged in (Duffecy et al., 2012) and number of messages posted (Classen et al., 2012). Jones et al. (2011) has used web metrics to classify users in online support forums. There are few studies that go beyond establishing a link with these types of components and analyzing the nature of what actually happened in that interaction. Results from some of the most well-known online Internet intervention research with cancer survivors suggests that use of the discussion board and social networking features was most related to increased learning in terms of the use of information services for cancer survivors (Shaw et al., 2007). In addition, in an online breast cancer support group, people who were more engaged were people who indicated a lack of competence in health communication, as well as having a lack of confidence in patient–physician communication (Han, Yoon, Sim, McTavish, & Gustafson, 2012), demonstrating a direct need for information.

Defining what is meant by engagement and analyzing components of what is defined as engagement and social networking is important and may lead to more beneficial programs. Peterson (2006) had defined engagement as “an estimate of the degree and depth of visitor interaction on the site against a clearly defined set of goals.” We would add to this definition by saying that engagement is an estimate of the degree, type, and depth of visitor interaction against a clearly defined set of goals.

The final two subsections of this background section attend to context for the participant population: (a) cancer survivors and (b) the intervention.

The Burden of Cancer Survivorship

Because of earlier detection and improved screening and treatments, the number of cancer survivors is increasing. It is estimated that there are nearly 12 million U.S. survivors, with 60% of these older than 65 years. The most common cancers among survivors are female breast (22%), prostate (20%), colorectal (9%), and gynecologic (8%; National Cancer Institute, 2012). While the growing number of survivors is encouraging, surviving cancer brings about a new set of challenges. Survivors need to be ever watchful for signs of recurrence at the same time that they attempt to continue with their normal role functions, including an increased interest in adopting new health behaviors into their lifestyle (McCorkle et al., 2011), and learning to deal with the emotional consequences of having survived cancer. In addition, they may face one or many comorbid conditions associated with survivorship, such as memory loss or muscle atrophy. Many survivors are encumbered by economic problems which stem from high utilization of health care, and problems with finding or retaining employment posttreatment (Hewitt, Greenfield, & Stoval, 2006). This well-documented burden suggests a need for behavioral interventions. Self-management programs were thought to be one possible solution. These are discussed in the next section.

Online Self-Management Programs

Self-management has been defined as “the tasks that individuals must undertake to live well with one or more chronic conditions. These tasks include having the confidence to deal with medical management, role management and emotional management of their conditions” (Adams, Greiner, & Corrigan, 2004). Good self-management can help with the side effects, both physical and psychological, that result from surviving cancer.

In a collaboration between Stanford and the University of Hawaii Cancer Center, a self-management program for cancer survivors was developed and named “Cancer: Thriving and Surviving.” Approximately 350 people went through the initial randomized trial (Bantum et al., 2014), which was built on a 30-year history of chronic disease self-management programs (Lorig et al., 1999; Lorig et al., 2008; Lorig et al., 2010; Lorig, Ritter, Laurent, & Plant, 2006). These past programs have been shown to improve health behaviors, health status, and sometimes health care utilization (Center for Disease Control and Prevention, 2011).

Using data from seven online “Cancer: Thriving and Surviving” workshops, this article explores the themes of participant’s discussions and links aspects of participant engagement with future health behavior and health status outcomes. The 6-week workshop contains both educational material on cancer survivorship skills and a social networking aspect. Participants interact on four-threaded bulletin boards (for more details, see the “Intervention” section).

In examining how online engagement may affect future health behaviors and health status we chose to examine four aspects of engagement. They were chosen because of their potential relevance to the development and implementation of future interventions. We report on these exploratory studies: (a) extracting themes from posts, (b) linking program engagement (number of posts written and number of posts read) with outcomes, (c) linking action planning completion (skills mastery) to outcomes, and (d) exploring the effects of self-tailoring (allowing participants to tailor the intervention to their own needs) on outcomes.

Although these analyses were conducted with cancer survivors, we believe that our results can also inform health education interventions for participants with other concerns.

Method

The Intervention

“Cancer: Thriving and Surviving” is a 6-week, password-protected interactive workshop for approximately 20 participants who may log on as often as they wish. The workshop has four components (Gits, Ritter, Plant, & Lorig, 2013):

Learning Center, which contains new interactive didactic material each week. The content was determined by a literature review of the needs of cancer survivors as well as a focus group with cancer survivors. Learning Center content included teaching three core skills (action planning, problem solving, and decision making), as well as offering participants suggestions for designing individualized exercise programs. It also includes use of cognitive symptom management techniques for managing negative emotions such as anger, fear, and depression; an overview of medications; aspects of physician–patient and family–patient communication; healthy eating; fatigue, sleep and pain management; and dealing with changes in body image.

Discussion Center, which consists of four-threaded bulletin boards (action planning, problem solving, celebrations, and difficult emotions). Posts to the Discussion Center are prompted by asking each participant to post one question a week. In addition, participants can post to any board at any time.

My Tools is a section that contains links to other sites as well as tools such as medication logs, an opportunity to keep a journal, exercise logs, meal planners, and audio relaxation exercises.

Post Office is where participants can correspond one-on-one with other participants without their post being seen by others. All workshops are facilitated by two peer facilitators. Information about these facilitators and their effects on outcomes can be found elsewhere.

The workshops are based on self-efficacy theory and systematically use techniques to enhance efficacy (Bandura, 1997). These include skills mastery through weekly action plans, modeling by peer facilitators, and discussions among participants, reinterpretation of symptoms by describing them as having many causes, and social persuasion by having public disclosure of both action plans and action plan completion as well as having participants publically state problems and help each other. The program differs from most behavioral interventions as all participants receive the same content. However, each individual is encouraged to choose his or her own weekly behavioral goals (action plans). These goals do not come from a list but may be anything from exercise, to visiting a friend to learning new words in a foreign language. It is not aimed at any one or two specific behavior changes. Rather it offers the structure and support for each participant to identify his or her own goals and activities to meet individual needs. Participants are asked each week to post an action plan and report on their progress the following week. No guidance is offered about the content of these plans. We have named this structured, self-directed behavior change as self-tailoring (Lorig & Holman, 2003).

Participants

Participants were recruited both online from cancer-related web sites and groups as well as from public service announcements in magazines and the Cancer Registry at Tripler Army Medical Center. All were enrolled in a larger randomized study to determine the efficacy of an Internet-based self-management program for cancer survivors, and this study includes the analysis of data provided during that larger trial. Specifically, participants were treatment subjects in the first seven online workshops. To enter the study, participants had to have completed primary treatment (surgery, chemotherapy, or radiation) at least 1 month and not more than 5 years before entering the study. They could not have had any cancer recurrence before starting the 6-month study period. People with all types of cancer except minor skin cancers were eligible.

For the current analyses, we excluded participants who, during the 6-month study period, had recurrences or other noncontrollable significant life events (e.g., family members fell seriously ill, n = 10). The occurrence of these events was independent of the parameters we studied in the population (e.g., number of action plans made by each participant), and we reasoned that the effects of these events on outcomes such as depression could overshadow any effect that our study parameters might have had. From the remaining set of participants, we only considered those who had completed both baseline and 6-month questionnaires.

Our inclusion criteria led to a total of 127 participants eligible for these studies (see Table 1 for participant demographics).

Table 1.

Participant Demographics.

Age in years, mean (range)	52 (26-81)
Gender, %
Female	81.8
Male	18.2
Education, %
High school	7.0
College	51.7
Graduate	36.0
Postgraduate	5.3
Months since diagnosis, mean (range)	29 (7-69)
Months since treatment, mean (range)	20 (3-53)

Outcome Measures

Outcome measures included depression, illness intrusiveness, stress, aerobic exercise, self-rated health, and visits to physicians in the past 6 months. Depression was measured by the PHQ-8 Scale (Kroenke, Spitzer, & Williams, 2001). Illness Intrusiveness (role function) consisted of 13 questions measuring the impact of disease on daily life. It was computed as the mean of five subscales, each of which is the mean of two or three of the items (Devins et al., 1983). Stress was measured using the Impact of Events Stress Scale (Horowitz, Wilner, & Alvarez, 1979). Aerobic exercise was reported as the number of minutes per day of moderate and vigorous leisure-time physical activity (MVLPA; Godin & Shephard, 1985; Godin, Valois, Shephard, & Desharnais, 1987). Reliability and validity of this measure is comparable with nine other self-report measures of exercise (Jacob, Ainsworth, Hartman, & Leon, 1993). Self-rated health was measured with a scale from the National Health Interview Survey (U.S. Bureau of the Census, 1985; Ware, Nelson, Sherbourne, & Stewart, 1992). Health care utilization over the prior 6 months was measured by self-report. In a study comparing the validity of self-reporting with chart audit, there were no biases toward improved reporting over time (Ritter et al., 2001).

Measuring Significance

We used permutation tests to measure significance of all hypotheses comparing a group of interest, that is, those who completed five or more action plans, with the group as a whole.

A permutation test compares changes observed in the group of interest (say, G) with changes in multiple sample populations of the same size drawn randomly from the whole population. The reported significance value reflects how likely is the observed change in our group of interest, G, compared with changes observed in randomly drawn groups of similar size. This is a conservative test as the comparison permutation groups include members of the group of interest. If the null hypothesis is upheld, then the change observed in the group of interest should not be significantly different from that observed in the set of randomly sampled populations. For each permutation test we drew subsets at random and without replacement from the overall population. For each permutation test drawn, we counted the proportion of subsets where the difference in the test statistic (e.g., mean change in depression scores from baseline to 6 months) was at least as large as the observed test statistic in the group of interest. The p values we report are thus the probability of seeing in the total population a difference at least as large as the difference observed in the population of interest.

There are reasons to be concerned about both Type 1 and Type 2 errors. The sample size is small leading to possible Type 1 errors. The p values are unadjusted for multiple hypothesis testing. The may lead to Type 2 errors. Because these are hypothesis-generating studies, associations with low p values between engagement behavior and downstream health outcomes serve as suggested hypotheses for further study.

The 127 participants in this study generated a total of 7,126 posts. These became the engagement data (corpus). Participants had a mean 15.6 (SD = 7.62) new discussion posts, 31.8 (SD = 35.1) replies to posts, and 137.4 (SD = 163.5) posts read. Mean number of “posts read” was based on visits to pages containing posts. Each original post was followed by 1 to 10 replies. Every time participants logged on they were informed if there were new posts or replies that they had not read and where to find these. Thus, participants could return to the same page several times. This was usually to view a new post which had been added to that page.

The Four Questions and Analyses, Purpose, Methods, and Results

In the following sections, we will discuss each of the four analyses sequentially, including further details of their purpose and method as well as the results.

Question 1: What Do Cancer Survivors Talk About? Themes From Posts

Purpose

The purpose of Analysis 1 was to understand the concerns shared by cancer survivors in an online workshop.

Methods

Themes were extracted using quantitative computational linguistics, which

Relies on an automated approach that allows analysis of a larger corpus (in our case, we analyzed 7,126 posts). Large qualitative data sets are nearly impossible to review qualitatively, and computational linguistics has been demonstrated in the literature to help gain insight into large document corpora (e.g., Griffths & Steyvers, 2004).

Can be less prone to subjective bias as suggested by the Association for Qualitative Research (2014). The strength and weakness of qualitative research is the human factor as the qualitative researcher is part of the process and this may include bias.

We used latent Dirichlet allocation (LDA) to analyze the participants’ posts (Blei, Ng, & Jordan, 2003). LDA is a method for discovering clusters of words that frequently co-occur (referred to as “topics” or “themes”) in large collections of documents. Intuitively, LDA seeks to identify common themes discussed in a selection of text. LDA is a probabilistic model for recovering the underlying structure of collections of text documents, and has been widely used for analyzing natural language documents within a variety of domains; for a compelling example, see Griffiths and Steyvers (2004), who use LDA on publications from 1991 to 2001 in the Proceedings of the National Academy of Sciences of the USA to extract scientific themes and topics that emerged during that decade. For a more detailed nontechnical explanation of LDA, see Blei, Ng, and Jordan (2003). LDA extracts groups of words that frequently co-occur in the text collection as “topics.” Each topic specifies a probability distribution over words; higher probability words are more likely to appear in posts containing that topic. Many but not all extracted topics map to cognitive themes and multiple topics may map to a single cognitive theme. Each text entity (e.g., a post) is modeled as words generated from a collection of topics associated with that post. LDA uses a statistical algorithm based on Monte Carlo inference. The model infers the collection of topics and their frequency in the corpus from the data. For each post, it infers the topics associated with that post and their prevalence within the post (Blei, Ng, & Jordan, 2003). In Table 2, we give an example of one post from this study and the corresponding top three topics mapped for this post by LDA. C-language code for LDA is available for free download and use from http://www.cs.princeton.edu/~blei/lda-c/index.html. That site also includes links to other implementations of LDA.

Table 2.

An Example Output of Topics Extracted From a Post Using Latent Dirichlet Allocation (LDA).

Post	“I have been really good about walking 30-40 mins almost every day ever since I read that it cuts the risk of breast cancer recurrence in half! That makes it pretty easy to stay motivated and put on my shoes to go walk. And I think the fact that I am not setting unattainable goals helps me stick to it as well. But, as of yet, I haven’t found the motivation to “kick it up a notch” to go beyond moderate paced walking or biking.”
Topic A	exercise work exercises find start gym routine fit exercising fitness workout hard walking schedule strength motivation working trainer stretching
Topic B	walk walking run day running good minutes weather mile treadmill back min started enjoy time helps days fast outdoors
Topic C	cancer breast treatment diagnosed risk years understand recurrence oncologist type negative cancers dx treatments cure diabetes research triple doctors

Note. The same word or words may appear in several topics.

For all our analyses, we ran LDA on the entire collection (corpus) of participant posts (action planning, celebrations, difficult emotions, and problem solving, along with replies to these posts). LDA requires specification of the number of topics to be extracted. We experimented with setting this number at 20, 40, 60, 80, and 100 topics. At 100 topics, the discovered topics were consistent between different runs of the algorithm and thus we report results from this set (reporting the most frequent topics). We also display the words associated with each topic (see Table 3). Topic names (cognitive themes) were determined by having three persons familiar with cancer survivorship read the words of each topic and independently name the topic. When there was a disagreement on the name, the senior reader chose the name.

Table 3.

Most Frequent Themes of Discussion in User Posts Extracted Using Latent Dirichlet Allocation.

Topic (collection of words)	Expert interpretation
good luck great idea make helps sounds hope nice suggestions agree job find check helpful point suggestion follow pretty	Encouragement (3.56%)
don’t feel time make things thing hard husband remember people bad put find understand realize stuff stay doesn’t love	Feeling guilty (3.46%)
time things good work give part life lot day long family spend exercise making enjoy thing taking step forward	Healing/Getting back to normal (3.24%)
cancer feel worry things time deal issues body normal fear feeling hard recurrence thoughts found anxiety mind pain big	Fear of recurrence (2.90%)
feel good makes don feeling make thing sense lot back bit body made hope helped strong sound depressed	Feeling better about self (2.88%)
great hear wonderful news hope happy glad love wow good sounds congratulations sharing amazing feeling bet congrats enjoy awesome	Congratulations (2.75%)
things years back thought find found past real made mind makes lot easier haven simple thing stop	Mixed emotions (2.42%)
life day live things living don change cancer today healthy future control enjoy guess moment good tomorrow worry thing	Living for the moment (2.16%)
friends family friend husband understand times hard talk life care feelings made support close person supportive thought live talked	Family/friends (2.14%)
plan action week forward good congratulations posting hearing plans specific completion sounds completing great tricia start success congrats completed	Action plan comments from facilitators (2.13%)
exercise work exercises find start gym routine fit exercising fitness workout hard walking schedule strength motivation working trainer stretching	Motivation for exercise (2.13%)
walk walking run day running good minutes weather mile treadmill back min started enjoy time helps days fast outdoors	Walking/running (1.83%)
didn’t people cancer told don treatment felt couldn’t life chemo wanted asked fact experience wasn’t thought found stuff bad	Insensitive friends/people (1.82%)
workshop support learned positive group hope share experience helped emotions continue made glad tools deal lot difficult great care	Thanks for support (1.81%)
plan week action good great job completing hope enjoy weeks congratulations walking luck vicki sounds fun planning tricia writing	Encouragement (action plan; 1.80%)
days day week times time exercise minutes walk goal great make enjoy vicci mind habit weeks easier left back	Unnamed (1.79%)
plan action peggy session week specific days make tools modify exercise glad sounds accomplishing complete work great planning peggyr	Action plan comments from facilitators (1.67%)
plan action confidence days week level tools plans make room specific high great wiggle change number successful increase decide	Confidence levels about action plans (1.63%)
years ago time cancer long year heart don died friend sister hear back hard months diagnosed looked father passed	Death/loss of loved ones to cancer (1.63%)

Note. Percentages listed in brackets represent the percentage of posts that discussed that theme.

We also report themes for two additional data sets: (a) the first post of each discussion thread (excluding replies from other participants) and (b) the first problem-solving discussion posts. These posts were in response to the prompt “Pick a cancer-related problem that you would like other workshop participants to offer ideas about” (see Table 4).

Table 4.

Frequent Themes of Discussion in User Posts Extracted Using Latent Dirichlet Allocation From (a) the First Post of All Discussion Threads (Table 4.1) and (b) the First Problem-Solving Post (Table 4.2).

4.1: Most prevalent original posts^a
“work,” “time,” “day,” “home,” “hours,” “tired,” “early,” “difficult,” “morning”	Work (4.78%)
“cancer,” “fear,” “life,” “worry,” “issues,” “health,” “scared,” “recurrence,” “diagnosis”	Cancer and recurrence (3.52%)
“exercise,” “gym,” “routine,” “fit,” “workout,” “hard,” “walking,” “schedule,” “strength,” “motivation”	Exercise (3.03%)
“beautiful,” “garden,” “fall,” “sun,” “watching,” “spring,” “birds,” “flowers,” “snow,” “summer”	Beautiful scenery (2.45%)
“friends,” “family,” “husband,” “understand,” “talk,” “care,” “feelings,” “support,” “close”	Emotional support (2.30%)
“pain,” “feet,” “neuropathy,” “chemo,” “toes,” “joints,” “tingling,” “hands,” “numbness,” “acupuncture”	Side effects and pain (2.28%)
“walk,” “run,” “day,” “good,” “minutes,” “weather,” “mile,” “treadmill,” “fast,” “outdoors”	Walking (2.02%)
“cancer,” “treatment,” “survivor,” “diagnosis,” “breast,” “stage,” “remission,” “free,” “fight”	Cancer (1.85%)
“scan,” “doctor,” “oncologist,” “test,” “results,” “mri,” “appointments,” “mammogram,” “blood”	Doctors and scans (1.80%)
“life,” “day,” “things,” “change,” “today,” “healthy,” “future,” “control,” “enjoy,” “moment,” “tomorrow,” “worry”	Life outlook (1.73%)
“workshop,” “support,” “learned,” “positive,” “group,” “share,” “experience,” “helped,” “emotions”	Workshop related (1.57%)
4.2: First problem-solving discussion posts
“chemo,” “brain,” “surgery,” “radiation,” “treatments,” “finished,” “remember,” “memory”	Chemo brain (3.09%)
“people,” “cancer,” “talk,” “answer,” “subject,” “survivor,” “share,” “avoid,” “conversation,” “questions,” “story,” “meet”	Talking to people (2.07%)
“weight,” “lose,” “pounds,” “gain,” “eating,” “control,” “exercise,” “diet,” “fat”	Losing weight (1.89%)
“sleep,” “bed,” “night,” “time,” “problems,” “late,” “hours,” “earlier,” “morning,” “trouble,” “asleep,” “stay”	Insomnia (1.81%)
“scars,” “surgery,” “mastectomy,” “breast,” “reconstruction,” “tattoo,” “courage,” “wear,” “clothes,” “proud,” “badge”	Scars (1.65%)

Note. Percentages listed in parentheses represent the percentage of posts that discussed that theme.

Original posts: Replies to post are excluded. Topics in boldface are those that were not among the top 20 topics in total posts shown in Table 3. ^bTopics found in top 20 topics from the first problem-solving posts and that were not in Table 3 or Table 4.1 categories.

Results

LDA generated 19 cognitive themes that occurred in at least 1% of the posts. These are shown in order by prevalence in all discussion posts in Table 3, along with an expert interpretation of themes. These themes represent nearly 40% of all posted content. When we examined only original posts, excluding replies to posts, new cognitive themes emerged concerning side effects/pain, cancer, and doctors/scans (shown in boldface in Table 4.1).

In addition, we examined the topic content and distribution for the first problem-solving discussion post. The most frequent cognitive themes in this collection of posts were broadly similar to those reported for all original posts (Table 4.2). Five new topics emerged; chemo brain, talking to people, losing weight, insomnia, and scars.

We analyzed prevalence of individual themes with changes in outcomes using permutation tests. After correcting for multiple hypotheses testing, no individual theme emerged as significant. This suggests that the participants’ experience and nature of engagement is multifactorial. There is not a single topic or type of interaction (e.g., support) that individually determines changes in outcome.

Question 2: Are Posting and Reading Posts Associated With Improvements in Health Behaviors and Health Status?

Purpose

Analysis 2 was conducted to explore the association between posting and reading behaviors and baseline to 6-month changes in health behaviors and health status. Posting and reading behavior are used as surrogates for measuring the level of program participation, making the assumption that those who post actively and/or read more are more engaged than people who post or read less.

Methods

To ascertain participant involvement in the workshop, we measured the number of new posts each participant initiated (not including replies to other posts), as well as the number of posts each participant read. Data came from the bulletin boards and the website’s event logs.

To determine if posting was associated with participant health behaviors or health status, we tagged the top 33% (n = 42) readers as belonging to a “top readers” group, and similarly we tagged the top 33% (n = 42) posters as “top posters.” Permutation tests as described above were used to determine the significance of the observed association between changes in outcome and the degree of program involvement.

Results: Effect of Involvement on Health Outcomes

In Table 5.1, we report for the population, the mean of the outcomes at baseline, at 6 months, and changes in the mean of outcomes in those 6 months. In Table 5.2, we see that for the group of top readers (n = 42), the change in outcomes over the 6-month period were similar to that of the population. This group had significantly higher baseline levels of illness intrusiveness and stress, both of which remained higher than the group as a whole at 6 months.

Table 5.

Engagement Measures and Their Association With Health Outcomes.

Engagement Behavior Subgroup	PHQ-8 Depression (0-24) ↓	Illness Intrusiveness (7-97) ↓	Stress (0-10) ↓	Exercise (min/week) ↑	General Health (1-5) ↓	MD Visits
5.1: Overall population (N = 127)
Baseline	6.13	2.73	21.6	238.8	2.43	3.95
6 months	5.43	2.42	20.2	294.7	2.39	3.88
Change	−0.70	−0.31	−1.4	55.9	−0.04	−0.07
5.2: Top 33% readers (n = 42)
Baseline	6.69	3.09 (.01)	24.0	247.4	2.50	4.30
6 months	5.85	2.63	22.6	322.6	2.48	4.55
Change	−0.84	−0.46	−1.4	75.2	−0.02	0.25
5.3: Top 33% posters (n = 42)
Baseline	7.29 (.04)	3.23 (.001)	26.0 (.002)	259.3	2.62	5.21 (.02)
6 months	6.24	2.71 (.04)	24.0 (.01)	359.5 (.03)	2.52	5.98 (.002)
Change	−1.05	–0.52 (.04)	−2.0	100.2 (.04)	−0.10	0.76
5.4: 5+ Action plans (n = 68)
Baseline	6.23	2.78	22.0	262.8	2.44	4.25
6 months	5.48	2.47	21.7	344.7 (.01)	2.43	4.79
Change	−0.75	−0.32	−0.3	81.9	−0.01	0.54
5.5: 5+ Successful action plans(n = 30)
Baseline	5.10	2.66	20.8	266.0	2.30	4.23
6 months	5.00	2.53	20.8	380.7 (.04)	2.30	4.33
Change	−0.10	−0.13	0.0	114.7 (.04)	0.00	0.10
5.6: 2+ Action plans (n = 110)
Baseline	6.10	2.77	20.8	252.6 (.01)	2.42	3.98
6 months	5.22	2.42	20.0	313.2 (.01)	2.37	3.97
Change	−0.88	−0.35	−0.8	60.6	−0.05	−0.01
5.7: 2+ Action plans, with >50% failed (n = 25)
Baseline	6.64	2.66	21.4	263.6	2.28	3.32
6 months	5.60	2.36	18.9	262.4	2.24	2.92
Change	−1.04	−0.30	−2.5	–1.2 (.05)	−0.04	−0.40
5.8: Self-tailoring among the Difficult-Emotions cohort (n = 40)
Baseline	7.70 (.01)	3.11 (.01)	25.9 (.004)	216.0	2.47	4.80
6 months	5.45	2.58	24.4 (.008)	225.0 (.02)	2.43	4.17
Change	–2.25 (.02)	–0.53 (.04)	−1.5	9.0 (.04)	−0.04	−0.63

Note. Significance was measured using permutation tests. Statistically significant deviations (p < .05) from the overall population are in boldface, with p values in parentheses. Ranges are given in parentheses following each outcome variable (top row). Arrows indicate whether lower (↓) or higher scores (↑) are desirable.

On the other hand, the group of top posters (n = 42) shown in Table 5.3 demonstrated significant improvements in both role function (illness intrusiveness) and exercise over 6 months. Similar to top readers, at baseline top posters had greater stress and depression and poorer role function than the population as a whole. They also used more health care services.

Question 3: What Is the Effect of Action Planning?

Purpose

Action planning is one way of achieving skills mastery, a key component of enhancing self-efficacy. It has become a key part of many behavioral interventions. Action plans generally have many of the following components: (a) they are time limited, (b) they focus on a specific behavior done in a specific amount on specific days or times, (c) they have a confidence level for completion, and (d) they include an opportunity for the individual to report back on plan completion. In this analysis, we examined how two of the key components of action planning (making plans and action plan completion) were related to future behavior change.

Methods

Each week within the 6-week workshop, participants were asked to make an action plan for the following week. At the first log-in of the next week they were asked if they fully completed, partially completed, or did not complete their plan. The plans and the completion of the plans were all entered on the action planning board in the Discussion Center. Data for this analysis were taken from the Action Plan Discussion Board. These data were also included in Analyses 1 and 2.

Each action plan was coded manually for a specific behavior, for example, exercise, changing diet, and so on. In addition, the participant’s responses to their action plans were coded as completed or not completed. Partial completion was coded as completed.

To analyze whether action plan formation and successful completion of these plans had an effect on health behaviors or health outcomes, we formed four overlapping groups of participants: Group 1 comprised participants who were very active in plan formation regardless of whether each plan was successful (5+ action plans out of a possible 6, n = 68). Group 2 comprised participants who were very successful in completing action plans (5+ successful action plans, n = 30). Group 3 comprised all participants who engaged in action planning (2+ action plans regardless of plan status, n = 110), and Group 4 comprised participants who formed action plans but often failed (2+ action plans with more than 50% failed completions, n = 25).

As in the analysis with the top posters and top readers group, we measured the differences in outcomes between these four groups and the overall population. Permutation tests were used to determine significant differences.

Results

Exercise was the dominant theme in the action plans, appearing in 56% of the action plan posts. Behaviors related to diet was the next most frequent action plan behavior (8%). No other action planning behavior accounted for more than 2%. We report the mean outcomes at baseline, at 6 months, and the mean change in outcome over those 6 months (Tables 5.5-5.7).

Group 1: At baseline, those who posted five or more action plans (Table 5.4) were similar to the overall population (Table 5.1). At 6 months they had increased their exercise by an average of 50 minutes more per week than the overall population.

Group 2: At 6 months, the group of successful action planning completers (Table 5.5) was exercising significantly more minutes per week (380 minutes compared with 295 minutes) than the overall population and also demonstrated a significant increase in exercise (115 minutes compared with 56 minutes).

Group 3: Participants who posted 2 or more action plans had similar outcomes to the group of active posters (Table 5.6).

Group 4: Participants who were unsuccessful in completing 50% or more of their action plans demonstrated a slightly negative change in weekly exercise (−1.1 minutes). This was significantly different then the 6-month change for the overall population (Table 5.7). There were no significant differences for any of the other health behaviors or health status variables between any of the groups and the overall population.

Fifty percent of the action plans involved exercise (Lorig, Laurent, Plant, Krishnan, & Ritter, 2013). It is therefore not surprising that of all the outcome variables, exercise was affected the most. The size of participant groups aggregated by non–exercise-related action plan goals were too small to draw conclusions about whether action planning was associated with change in corresponding outcome.

Question 4: What Is the Effect of Self-Tailoring?

Purpose

The purpose of this analysis was to determine if the salient problem as stated by the participant improves over 6 months compared with the improvement in the same outcome variable for the overall group.

Salient belief theory states that the first expression of beliefs or problems is of those that are most salient to the individual (Miller, 1956). Based on this theory, participants are asked at the time of registration and then again during their first post to share with the group a problem they have as a result of being a cancer survivor. Thus, problems are not assumed but are specific to each individual. In addition, the workshop does not urge all participants to change the same behaviors. Rather it is hypothesized that when participants actively state a problem it follows that the participant will choose tools and make action plans toward lessening this problem. They do this without the investigators manipulating the intervention. In other words they may self-tailor their activities as a way of lessening or solving their problem. For example, participants who voiced negative emotions such as depression or fear (a salient belief) in this first post, might be expected to choose activities that would improve these emotions and in turn lessen their depression. Although the problem may be the same, the activities chosen by the individual participants may be different.

It is important to differentiate self-tailoring from tailoring. Kreuter (2000) defined tailoring as “any combination of information or change in strategies intended to reach one specific person, based on characteristics that are unique to that person, related to the outcome of interest, and have been derived from an individual assessment.” Hawkins, Kreuter, Resnicow, Fishbein, and Dijkstra (2008) expanded on this. “Tailoring” means creating communications in which information about a given individual is used to determine what specific content he or she will receive, the contexts or frames surrounding the content, by whom it will be presented, and even through which channels it will be delivered.

These definitions clearly delineate two characteristics of tailored interventions. First, messages, means of delivery, and so on, are based on some known characteristics of the individual, and second, the messages, modes of communication, and so on, are created specifically for that individual based on the above knowledge.

For Cancer: Thriving and Surviving, neither the messages nor modes of delivery meet either of these criterion. Knowledge about individual participants is not used to create messages, or delivery methods and all participants receive the same messages. Although it is true that some tailored interventions allow participants to make choices these serve as information about the individual to tailor messages, in Cancer: Thriving and Surviving, the individual makes choices about what he or she wishes to do and/or discusses in the online discussion center. The program is delivered exactly the same to all participants and is designed to facilitate the accomplishment of individually derived outcomes and or discussions. In this way, self-tailoring is more akin to goal attainment theory described by Kiresuk and Sherman (1968) as a method to evaluate mental health interventions. In short, tailoring has to do with content and its delivery. Sometimes with tailoring, participants are given a limited choice of behaviors. Self-tailoring has to do with each individual choosing behaviors related to the participant-specific problems. There are an unlimited possibility for both behaviors and problems.

An example of self-tailoring from another online study (Lorig et al., 2013) is a woman who week-after-week made an action plan to learn two new words of a Native American language. The facilitators never questioned this plan. After the program, we asked about the plan, as it was unusual. She responded that she was a cancer survivor who wanted to become more fluent in her language and had asked her doctor what to do for chemobrain. He suggested memory exercises. This was her solution.

Methods

To measure the effect of self-tailoring, we grouped the participants based on the problem identified in their first problem-solving post. Each post was manually coded with one or more of the participant outcomes under consideration (e.g., a post about being constantly depressed at the thought of recurrence is classified under “depression”). The set of possible labels comprised “depression,” “illness intrusiveness,” “stress,” “diet,” “fatigue,” and “exercise.” Ambiguous posts were separated and labeled by three different people, with the majority determining the final code. Depression comprised 31% of these posts. No other outcome code reached 20%. Thus, only the group that discussed depression as their salient problem had a large enough sample for analysis (n = 40). Hereon, we refer to this group as the difficult-emotion cohort. In assessing the benefit of self-tailoring on change in depression at 6 months, we should control for two factors: (a) the possibility of regression to the mean and (b) time since last cancer recurrence. Participants who have been diagnosed most recently are more likely to be depressed (Giese-Davis et al., 2011).

To control for the possibility of regression to the mean we used linear regression. This determined that each point of baseline PHQ-Depression was related, on average, to 0.4 points of improvement after 6 months (i.e., if person A started off 1 point more depressed than person B at baseline, on average, person A would improve over the next 6 months by 0.4 points relative to person B on average). Based on this, we corrected each person’s change in depression after 6 months to take into account his or her baseline depression.

To control for the time since treatment, we ran stratified permutation tests on the corrected depression values. These involved binning participants by their time since treatment (with bin sizes of 3 months), and then drawing random subsets in the permutation tests such that these random subsets had the same distribution of the time since treatment as the original subsets. The p values we report from these tests are thus the probability of seeing a difference at least as large as the observed difference, under the null hypothesis that only the time since treatment determines the difference. Using the corrected depression values, this test simultaneously controls for both baseline depression and time since treatment.

We also compared the scores for the difficult-emotions cohort to a cohort of the same number (n = 40) of those most depressed at baseline (see box-and-whiskers plot, Figure 1).

Figure 1.

Distribution of depression scores in the self-tailoring difficult-emotions cohort, and the control depression cohort at baseline (left) and after 6 months (middle), and the corresponding distribution of changes in depression scores from baseline to 6 months (right).

Results: Effect of Self-Tailoring on Health Outcomes

In Table 5.8, we report the mean at baseline, at 6 months, and the mean 6-month change in outcomes for all health outcomes (including depression) for the difficult-emotions cohort. These same data for the overall population are in Table 5.1. The difficult-emotions cohort had significantly more depressive symptoms at baseline, with a mean PHQ-8 score of 7.7 compared with the overall population’s mean of 6.1 (p = .01). Over 6 months, the difficult-emotions cohort improves by 1.6 points. This result is significant even when controlling for baseline depression and the time since treatment. (p = .034). In Figure 1, we show box-and-whisker plots of the depression scores of both this self-selected difficult-emotions cohort (n = 40), and a control cohort comprising the 40 participants who were not in the difficult-emotions cohort but had the highest depression scores at baseline. At baseline, these two cohorts have similar depression scores (Figure 1–left); however, after 6 months, the self-tailoring, difficult-emotions cohort shows markedly more improvement than the control cohort (Figure 1–mid). This difference is reflected in Figure 1–right, which summarizes the distribution of the individual changes in depression in both cohorts. These results demonstrate visually that the improvement of the self-tailoring difficult-emotions cohort is not solely due to its high levels of baseline depression.

Moreover, the difficult-emotions cohort had greater illness intrusiveness and stress compared with the overall population at baseline. While both depression and illness intrusiveness (mean decrease of 0.53) significantly improved after 6 months, stress remained higher than that of the population (24.4 vs. 20.2). Minutes of exercise per week significantly improved for the difficult-emotions cohort, but this change was not significant compared with the overall sample.

Discussion

The above sets of analyses represent an exploration of the type of posts made by cancer survivors to an online workshop as well at the effects of types of engagement, posting, reading posts, action planning, and self-tailoring on outcomes. These studies have several limitations. First, the overall sample may not be generalizable to populations outside cancer survivors or even those outside cancer survivors who post on the Internet. Second, the sample was small and in many cases made smaller by stratification. The few numbers of males and ethnic minorities resulted in us not looking at demographic differences in engagement. This is something that should be considered for future study. Finally, we conducted multiple tests. Any significant values must be considered exploratory and may be due to chance. At the same time, it may be that significance was limited by small sample size.

Despite these limitations, there are some intriguing findings.

In Analysis 1 (theme identification), one of the most interesting findings is that only one of the top themes was cancer specific. No doubt, other themes were discussed in the context of cancer survivorship. In addition, the most prevalent themes were positive (encouragement and positive affirmation). Facilitator support for action plans appeared twice in these themes with one theme involving helping with action plans whereas the second was congratulatory.

When we examined only posts originated by participants (excluding posts that were responding to other posts), three new themes emerged, all directly related to cancers: side effects, cancer, and doctors. Encouragement and positive affirmation disappeared from the list of top themes and work (or functioning in the well world) became the most frequent theme. Thus, we can begin to see the struggle of survivors as they move between the world of illness and the world of wellness. This suggests that in conducting theme analyses, it may be important to differentiate the type of communication (initiating as opposed to responding). It is likely that there are further differentiations within each of these categories. This was illustrated when we examined the first (salient) post of each participant (Question 4). Understanding the expectations participants feel or the sort of norms for this type of online communication are also important, as they could be driving some of the content of the posts.

Many of these themes: chemo brain, talking to people about cancer, weight loss, insomnia and scars, did not appear in either of the first two lists.

Taken as a whole, we might conclude that themes differ depending on if they are solicited, such as the first problem-solving post, are in response to others, such as reflected in the overall themes, or are themes originated by participants. It may also be that themes differ as to when they appear in the life of the social network, based, in part, on timing of specific Learning Center content or facilitator prompts. This was unexplored in the current work and should be conducted in future analyses. The importance of how and when a theme occurs as well as the demand characteristics of the online environment is open to further study.

In Analysis 2, we examined whether engagement (reading and/or posting) was related to outcomes. While both top readers and posters are more depressed than the overall population, it is interesting to note that active involvement and interaction in the form of posting is associated with larger improvement in outcomes. Specifically we found that those who posted the most also appeared to get the most benefit. This same benefit was not seen by those who read posts but were more reticent at posting. This may indicate that helping others (posting) is in and of itself an important intervention. This notion builds on an earlier study in which participants in another self-management intervention indicated that one of the most useful parts of the intervention was helping others (Campbell, Sengupta, Santos, & Lorig, 1995).

If this finding were to apply to other social networks it would suggest that the active posters, as opposed to the lurkers, gain the most benefit from these networks. This finding might be extended to small group and one-on-one communication but again more study is needed.

In Analysis 3, we examined the effects of making and completing action plans. The most frequent theme of an action plan was exercise. Thus, it would be expected that amount of exercise at 6 months post-baseline would be most affected by action-planning behavior. It appears that those who made and completed the highest number of action plans benefit most and that those who make action plans but fail to complete them were exercising less at 6 months. This finding has been shown in another recent study (Lorig et al., 2013) and has important implications for clinical practice (see next section). A caveat to this discussion is that the action planning findings may be confounded by the personality of the participants; motivated participants are likely to make and succeed on more action plans. External circumstances such as illness, family stress, or work pressures may have also confounded this result. We cannot entirely exclude the possibility that attempting action planning and failing may have deleterious results. While this will take more study, it should be used as a caution.

Analysis 4 formally introduced the concept of self-tailoring and presents early data that suggest allowing participants to freely choose both the problem they wish to work on and the way(s) in which they will address this problem may have positive results. Specifically, the participants who named feeling depressed as their salient problem were also those who had the most improvement in depression even after controlling for baseline depression and time since diagnosis. They did so by different routes that were guided, but not stipulated by, the program: Some people did more exercise, others reestablished old friendships, and still others used stress-reduction techniques. The common denominator was that the program provided an efficacy-enhancing structure in which these individuals could gain confidence in their abilities. Recently, Spring et al. (2012) have found that if people make a change in one health behavior, they tend to also change other behaviors. Thus, it may be that self-tailoring will affect improvements in multiple health behaviors. In a future a trial, comparing tailored with self-tailored interventions may be useful.

Implications for Practice

Although none of these findings are conclusive, some, based on this initial evidence and intuition, can be applied to practice.

In the theme analysis, the high number of themes related to the giving of support, by both the facilitators and the other group members, highlights the need to formally structure such support into our interventions. Because of time constraints, or the fear of misinformation being passed around, we often control participant interactions by asking questions or completing set exercises. It would appear that the free expression of support is an important component of this and possibly other interventions. An extensive literature exists on the importance of social support. The literature on offering support is much smaller. We may need to reframe our interventions so that participants are able to and encouraged to both give and receive support. This has been explored by Ingersoll-Dayton and Antonucci (1988), among others.

A potential drawback to the use of LDA for this study is its decontextualization. A future study of interest would be to compare LDA with more traditional qualitative techniques.

We have long known that interactive health education is usually better than passive health education. This finding was replicated in this study where “posters” demonstrated more improvements than “readers.” This suggests that group interventions should be planned to include active participation by the participants.

While it is not surprising that action planning is associated with positive behavior change, there is little literature suggesting that failing to complete multiple action plans is associated with poorer outcomes. For behavioral interventions and in clinical practice, this suggests that when assisting people with action plans it is important to both do everything possible to assure success and to follow up with completion status. It may be that those who continually fail to complete their action plans should not be asked to take part in this activity.

While there is a long history of and belief in tailored interventions, little is known about self-tailoring. The current study is an early demonstration that letting people work on their own problems in their own way within a structured and supportive framework may lead to improved outcomes. This is at least worth consideration when developing interventions.

A final implication for practice is the interdisciplinary nature of this work. The authors came from at least three different academic cultures, behavioral science, anthropology, and computational linguistics. We speak different scientific languages, use different tools, and know almost nothing about each other’s disciplines. Nevertheless, using both new and old tools, reading each other’s literature, and asking many questions, we have attempted to create beginning insights into some intriguing questions. We would urge others to take similar journeys into the unknown. This can only lead to improved practice.

Footnotes

Authors’ Note

Katy Plant, Diana Laurent, Angela Devlin, and Eileen Bradley helped with the implementation of the trial. Marc Rasi and Alexander Gits contributed to the early stages of this study. Gerald Devins assisted in annotating themes.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The U.S. Department of Defense (W81XWH-06-2-0042) and Stanford Cancer Center funded the original Cancer Survivors trial, which was the source of the data used for this study. This study was supported in part by the Amgen Foundation and the Stanford Cancer Center.

References

Adams

Greiner

A. C.

Corrigan

J. M.

(Eds.). (2004). Report of a summit. The First Annual Crossing the Quality Chasm Summit—A focus on communities. Washington, DC: National Academies Press.

The Association for Qualitative Research. (2014). Bias. Retrieved from http://www.aqr.org.uk/glossary/?term=bias

Bandura

(1997). Self-efficacy, the exercise of control. New York, NY: Freeman.

Bantum

E. O.

Albright

C. L.

White

K. K.

Berenberg

J. L.

Layi

Ritter

P. L.

. . . Lorig

(2014). Surviving and thriving with cancer using a web-based health behavior change intervention: Randomized controlled trial. Journal of Medical Internet Research, 16, e54. doi:10.2196/jmir.3020

Berry

D. S.

Pennebaker

J. W.

(1993). Nonverbal and verbal emotional expression and health. Psychotherapy and Psychosomatics, 59, 11-19. doi:10.1159/000288640

Blei

D. M.

A. Y.

Jordan

M. I.

(2003). Latent Dirichlet allocation. Journal of Machine Learning Research, 3, 993-1022. Retrieved from http://blog.echen.me/2011/08/22/introduction-to-latent-dirichlet-allocation/

Campbell

B. F.

Sengupta

Santos

Lorig

K. R.

(1995). Balanced incomplete block design: Description, case study, and implications for practice. Health Education and Behavior, 22, 201.

Centers for Disease Control and Prevention. (2011). Sorting through the evidence for the Arthritis Self-Management Program and the Chronic Disease Self-Management Program (Executive summary of the ASMP/CDSMP meta-analyses). Atlanta, GA: Author.

Classen

C. C.

Chivers

M. L.

Urowitz

Barbera

Wiljer

O’Rinn

Ferguson

S. E.

(2012). Psychosexual distress in women with gynecologic cancer: A feasibility study of an online support group. Psycho-Oncology. Advance online publication. doi:10.1002/pon.3058

10.

Devins

Binik

Hutchinson

Hollomby

D. J.

Barré

P. E.

Guttmann

R. D.

(1983). The emotional impact of end-stage renal disease: Importance of patients’ perceptions of intrusiveness and control. International Journal of Psychiatry in Medicine, 13, 327-343.

11.

Duffecy

Sanford

Wagner

Begale

Nawacki

Mohr

D. C.

(2012). Project onward: An innovative e-health intervention for cancer survivors. Psycho-Oncology. Advance online publication. doi:10.1002/pon.3075

12.

Giese-Davis

Collie

Rancourt

K. M.

Neri

Kraemer

H. C.

Spiegel

(2011). Decrease in depression symptoms is associated with longer survival in patients with metastatic breast cancer: A secondary analysis. Journal of Clinical Oncology, 29, 413-420.

13.

Gits

A. A.

Ritter

P. L.

Plant

Lorig

(2013). Coding and evaluating facilitator posts for an online cancer survivor workshop. Journal of Psychosocial Oncology. Advance online publication. doi:10.1080/07347332.2012.761321

14.

Godin

Shephard

(1985). A simple method to assess exercise behavior in the community. Canadian Journal of Applied Sport Sciences, 10, 141-146.

15.

Godin

Valois

Shephard

R. J.

Desharnais

(1987). Prediction of leisure-time exercise behavior: A path analysis (LISREL V) model. Journal of Behavioral Medicine, 10, 145-158.

16.

Gooden

R. J.

Winefield

H. R.

(2007). Breast and prostate cancer online discussion boards: A thematic analysis of gender differences and similarities. Journal of Health Psychology, 12, 103-114.

17.

Griffiths

T. L.

Steyvers

(2004). Finding scientific topics. Proceedings of the National Academy of Sciences, 101, 5228.

18.

Gustafson

D. H.

Hawkins

Pingree

McTavish

Arora

N. K.

Mendenhall

. . .Salner

(2001). Effect of computer support on younger women with breast cancer. Journal of General Internal Medicine, 16, 435-445. doi:10.1046/j.1525-1497.2001.016007435.x

19.

Han

Yoon

Sim

McTavish

F. M.

Gustafson

D. H.

(2012). Social and psychological determinants of levels of engagement with an online breast cancer support group: Posters, lurkers, and non-users. Journal of Health Communication, 17, 356-371.

20.

Hawkins

R. P.

Kreuter

M. W.

Resnicow

Fishbein

Dijkstra

(2008). Understanding tailoring in communicating about health. Health Education Research, 23, 454-466.

21.

Hewitt

M. E.

Greenfield

Stoval

(2006). From cancer patient to cancer survivor: Lost in transition. Washington, DC: National Academies Press.

22.

Horowitz

Wilner

Alvarez

(1979). Impact of event scale: A measure of subjective stress. Psychosomatic Medicine, 41, 209-218.

23.

Ingersoll-Dayton

Antonucci

T. C.

(1988). Reciprocal and nonreciprocal social support: Contrasting sides of intimate relationships. Journal of Gerontology, 43, S65-S73. doi:10.1093/geronj/43.3.S6

24.

Jacob

D. R.

Ainsworth

B. E.

Hartman

T. J.

Leon

A. S.

(1993). A simultaneous evaluation of 10 commonly used physical activity questionnaires. Medicine & Science in Sports & Exercise, 25, 81-91.

25.

Jones

Sharkey

Smithson

Ford

Emmens

Hewis

. . . Owens

(2011). Using metrics to describe the participative stances of members within discussion forums. Journal Medical Internet Research, 13, e3. doi:10.2196/jmir.1591

26.

Kiresuk

T. J.

Sherman

R. E.

(1968). Goal attainment scaling: A general method for evaluating comprehensive community mental health programs. Community Mental Health Journal, 4, 443-453.

27.

Klemm

(2008). Late effects of treatment for long-term cancer survivors: Qualitative analysis of an online support group. Computers, Informatics, Nursing, 26, 49-58.

28.

Kreuter

M. W.

(2000). Tailoring: What’s in a name? [Editorial]. Health Education Research, 15, 1-4.

29.

Kroenke

Spitzer

R. L.

Williams

J. B. W.

(2001). The PHQ-9. Journal of General Internal Medicine, 16, 606-613.

30.

Lieberman

(2007). The role of insightful disclosure in outcomes for women in peer-directed breast cancer groups: A replication study. Psycho-Oncology, 16, 961-964. doi:10.1002/pon.1181

31.

Lieberman

M. A.

Goldstein

B. A.

(2006). Not all negative emotions are equal: The role of emotional expression in online support groups for women with breast cancer. Psycho-Oncology, 15, 160-168.

32.

Lieberman

M. A.

Winzelberg

(2009). The relationship between religious expression and outcomes in online support groups: A partial replication. Computers in Human Behavior, 25, 690-694.

33.

Lorig

Laurent

D. D.

Plant

Krishnan

Ritter

P. L.

(2013). The components of action planning and their associations with behavior and health outcomes. Chronic Illness, 10, 50-59. doi:10.1177/1742395313495572

34.

Lorig

K. R.

Holman

H. R.

(2003). Self-management education: History, definition, outcomes, and mechanisms. Annals of Behavioral Medicine, 26, 1-7.

35.

Lorig

K. R.

Ritter

P. L.

Dost

Plant

Laurent

D. D.

McNeil

(2008). The Expert Patients Programme online, a 1-year study of an Internet-based self-management programme for people with long-term conditions. Chronic Illness, 4, 247-256.

36.

Lorig

K. R.

Ritter

P. L.

Laurent

D. D.

Plant

(2006). Internet-based chronic disease self-management: A randomized trial. Medical Care, 44, 964-971.

37.

Lorig

K. R.

Ritter

P. L.

Laurent

D. D.

Plant

Green

Jernigan

V. B. B.

Case

(2010). Online diabetes self-management program: A randomized study. Diabetes Care, 33, 1275-1281.

38.

Lorig

K. R.

Sobel

D. S.

Stewart

A. L.

Brown

B. W.

Bandura

Ritter

. . . Holman

H. R.

(1999). Evidence suggesting that a chronic disease self-management program can improve health status while reducing hospitalization: A randomized trial. Medical Care, 37, 5-14.

39.

McCorkle

Ercolano

Lazenby

Schulman-Green

Schilling

L. S.

Lorig

Wagner

E. H.

(2011). Self-management: Enabling and empowering patients living with cancer as a chronic illness. CA: A Cancer Journal for Clinicians, 61. 50-62. doi:10.3322/caac.20093

40.

Miller

G. A.

(1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81-97.

41.

P. K. H.

Malik

S. H.

Coulson

N. S.

(2009). Gender differences in computer-mediated communication: A systematic literature review of online health-related support groups. Patient Education and Counseling, 75, 16-24. doi:10.1016/j.pec.2008.08.029

42.

National Cancer Institute, Office of Cancer Survivorship. (2012). Statistics. Retrieved from http://cancercontrol.cancer.gov/ocs/statistics/statistics.html

43.

Peterson

E. T.

(2006, December 18). How do you calculate engagement? Part II. Web analytics demystified. Retrieved from http://blog.webanalyticsdemystified.com/weblog/2006/12/how-do-you-calculate-engagement-part-ii.html

44.

Ritter

P. L.

Stewart

A. L.

Kaymaz

Sobel

D. S.

Bloch

D. A.

Lorig

K. R.

(2001). Self-reports of health care utilization compared to provider records. Journal of Clinical Epidemiology, 54, 136-141.

45.

Shaw

Han

J. Y.

Kim

Gustafson

Hawkins

Cleary

. . . Lumpkins

(2006). Effects of prayer and religious expression within computer support groups on women with breast cancer. Psycho-Oncology, 16, 676-687. doi:10.1002/pon.1129

46.

Shaw

B. R.

Han

J. Y.

Baker

Witherly

Hawkins

R. P.

McTavish

Gustafson

D. H.

(2007). How women with breast cancer learn using interactive cancer communication systems. Health Education Research, 22, 108-119.

47.

Spring

Schneider

McFadden

Vaughn

Kozak

A. T.

Smith

. . . Lloyd-Jones

D. M.

(2012). Multiple behavior changes in diet and activity; a randomized controlled trial using mobile technology. Archives of Internal Medicine, 172, 789-796.

48.

U.S. Bureau of the Census. (1985). National Health Interview Survey. Washington DC: U.S. Department of Commerce.

49.

Ware

J. E.

Jr. Nelson

E. C.

Sherbourne

C. D.

Stewart

A. L.

(1992). Preliminary tests of a 6-item general health survey: A patient application. In Stewart

A. L.

Ware

J. E.

(Eds.), Measuring functioning and well-being: The medical outcomes study approach (pp. 291-303). Durham, NC: Duke University Press.

50.

Winzelberg

A. J.

Classen

Alpers

G. W.

Roberts

Koopman

Adams

R. E.

. . . Taylor

C. B.

(2003). Evaluation of an Internet support group for women with primary breast cancer. Cancer, 97, 1164-1173.

51.

Wise

Han

J. Y.

Shaw

McTavish

Gustafson

D. H.

(2008). Effects of using online narrative and didactic information on healthcare participation for breast cancer patients. Patient Education and Counseling, 70, 348-356.