ComputerLink was a computer support network for family caregivers of people with Alzheimer's disease. In a 12-month experiment, 102 caregivers were randomly assigned to an experimental group that had access to ComputerLink or to a control group that did not. This investigation examined whether caregivers in the experimental group had greater reductions in four types of care-related strain by the end of the 1-year study. ComputerLink reduced certain types of strain if caregivers also had larger informal support networks, were spouses, or did not live alone with their care receivers. More frequent use of the communication function was related to significantly reduced strain for caregivers who were initially more stressed and for non-spouse caregivers. Greater use of the information function was related to significantly lower strain among caregivers who lived alone with care receivers. Overall, ComputerLink appeared to be an effective tool for reducing strain for some caregivers.

This investigation examined whether a computer support network for family caregivers of people with Alzheimer's disease reduced levels of care-related strain. The computer support network was named ComputerLink and involved 102 primary informal caregivers. Half of these caregivers were randomly assigned to an experimental group that could access ComputerLink. The remaining caregivers served as a control group and did not receive a computer or the ComputerLink software. The experiment lasted 12 months and included extensive pre- and posttest interviews, ongoing monitoring of ComputerLink use, and monthly monitoring of formal service use. Findings reported here compared levels of four types of caregiver strain for experimental and control group subjects and examined variation in strain associated with different amounts of ComputerLink use within the experimental group.

ComputerLink System

Caregivers in the experimental group received a Wyse 30 terminal and 1200 baud modem. Although outdated by current standards, et the time of the experiment (1990), this hardware was an economic and workable option. The applications available in ComputerLink were similar to comparable systems currently in use (Slatalla, 1996). Terminals were installed in caregivers' homes by project nurses, who also provided training.

The most widely used feature of ComputerLink was the communication component, which enabled caregivers to interact over the system. It included a public bulletin board, private mail, and a question-and-answer segment facilitated by a nurse. Another feature, which was a solitary component, was a four-module electronic encyclopedia on Alzheimer's disease and its treatment, management of symptoms, services for Alzheimer's patients and caregivers, and self-care for caregivers. The content of the encyclopedia was developed in collaboration with the Cleveland Area Chapter of the Alzheimer's Association. The solitary component also included a decision-support module that helped caregivers address unsolved problems and dilemmas.

Forty-seven of the 51 experimental group caregivers completed the intervention and logged into ComputerLink 3,937 times over the 12-month project, an average of 6.98 times per month. Nearly one fourth of caregivers logged into the system 10 or more times per month, whereas another one fourth logged in only once or twice per month. Caregivers spent an average 1.6 hours per month on the system, with the heaviest users spending 4 to 6 hours per month (see Brennan, Moore, & Smyth, 1995, for a full description of the ComputerLink system).

ComputerLink as a Support Service

ComputerLink was conceived as an alternative to traditional caregiver support services, such as support groups and health education programs, and is similar to other systems for caregivers and people dealing with other health or social problems (Alemi et al., 1996; Brown-Ewinget al., 1988; Christ & Siegel, 1990; Fernsler & Manchester, 1997; Gustafson et al., 1994; Gustafson, Taylor, Thompson, & Chesney, 1993; Paperny & Starn, 1989; Robinson, 1989). ComputerLink differs from computer networks such as SeniorNet (Greenberger & Puffer, 1989) because of its exclusive focus on managing family caregiving. ComputerLink was expected to be a more convenient and effective intervention than traditional support services because (a) participation did not require caregivers to leave home, (b) it could be accessed any time day or night, (c) use could be anonymous or self-identified, and (d) information could be tailored to individual needs (Schmall, 1984 Smyth & Harris, 1993; Wright, Lund, Pett, & Caserta, 1987).

Despite the potential advantages of a computer support network, few studies have tested whether such a network has the intended effects of providing support and assistance while reducing levels of caregiver strain, which is the outcome of interest in this investigation (Alemi & Stephens, 1996; Smyth & Harris, 1993). Related research on the effects of conventional types of caregiver support services have produced inconsistent evidence that caregiver strain is reduced by supportive interventions (for reviews, see Knight, Lutzky, & Macofsky-Urban, 1993; Toseland & Rossiter, 1989; Zarit & Teri, 1991). A definitive conclusion about the utility of caregiver support services is hampered by three methodological limitations of most existing studies (Kosloski & Montgomery, 1995; Montgomery, 1996; Whitlatch, Zarit, & von Eye, 1991).

First, there often is a lack of uniform exposure to the intervention among experimental subjects and an inability to control exposure to similar interventions among controls (Knight et al., 1993; Lawton, Brody, & Saperstein, 1991; Montgomery & Borgatta, 1989). Kosloski and Montgomery (1995) illustrated the importance of this issue when they found a previously undetected treatment effect, once experimental group caregivers who did not use an intervention were dropped from the analysis. In the ComputerLink experiment, this limitation was addressed by directly testing for differences in caregiver strain associated with different amounts and types of exposure to ComputerLink within the experimental group. Because ComputerLink was available only to the experimental group and few other computer networks were available at the time of the study, control group caregivers were not exposed to similar interventions.

A second limitation is that a single supportive intervention often is composed of a complex and diverse set of components. Even when the overall amount of exposure is the same, the use of various components may differ, and different components may have different effects (Bass, Noelker, & Rechlin, 1996; Bourgeois, Schulz, & Burgio, 1995).To address this potential limitation, the ComputerLink experiment employed a passive monitoring system that recorded the exact component of the system used by caregivers and enabled different components to be distinguished in the analysis. Two distinct components were examined in this study: (a) the communication component, which involved interacting with other caregivers, and (b) the solitary component, which included individualized learning and decision support.

A third limitation of many prior evaluations is that the effects of an intervention are assumed to be uniform for all those who have access. An alternative approach corresponds to the buffering hypothesis, which posits that the effect of an intervention will vary depending on caregivers' initial levels of stress and vulnerability (Lin, 1986; Whitlatch et al., 1991). The beneficial consequences of an intervention will not be uniform; rather, caregivers who are initially more stressed or vulnerable will experience greater benefits. The buffering effect is modeled by a statistical interaction between initial stress and use of a supportive intervention (Finney, Mitchell, Cronkite, & Moos, 1984; Southwood, 1978).

The buffering model, depicted in Figure 1, guides this evaluation. The first part of the model includes measures of four types of initial (T1) caregiver strain (physical, emotional, and relationship strain, as well as activity restriction) and three characteristics that heighten vulnerability to care-related stress (whether the caregiver was the spouse of the care receiver, a lack of informal support, and being the only person living with the care receiver). In accord with the buffering model, it is hypothesized that caregivers with heightened initial levels of stress and those with heightened vulnerability to stress will experience greater benefits from ComputerLink, as reflected in greater reductions in care-related strain at the end of the 1-year study period.

The T1 measures of the four dimensions of caregiver strain directly assess negative consequences of caregiving at the time the experiment began. The other three constructs are risk factors that elevate the potential for negative caregiving consequences. Spouses, especially wives, are at greater risk because they tend to be older, have more health problems and vulnerabilities, become more isolated in the caregiving role, and are most committed to keeping their husbands or wives at home regardless of care-related difficulties (Cantor, 1982 Cicirelli, 1983; George & Gwyther, 1986; Miller, McFall, & Montgomery, 1991; Wright, Clipp, & George, 1993). Lack of informal support represents the number of family members and friends in the caregiver's support network. Caregivers with more family and friends have more resources to overcome care-related difficulties and are more likely to get affective and instrumental assistance with care (Clipp & George, 1990; George & Gwyther, 1986). Conversely, caregivers who lack network of family and friends were considered more vulnerable because they have fewer sources of support and assistance. Similarly, the measure representing whether the caregiver was the only person residing with the care receiver reflects potential support from other household members. Caregivers living alone with care receivers are at greater risk because they are the only proximate provider of care and are less likely to have a secondary helper who can provide on-site relief and emotional support.

The next part of Figure 1 is analogous to social support in the buffering model and includes three measures of the ComputerLink intervention. The first measure, ComputerLink treatment, distinguishes between the experimental and control groups. The second measure is only for the experimental group and represents the frequency of use of the communication function to interact with other caregivers. The third measure also is only for the experimental group and represents the frequency of use of the solitary function for individualized information and decision making. The measures of T1 care-related stress and ComputerLink use are expected to interact in the buffering model, as indicated by the vertical arrow in Figure 1, which intersects the path between care-related stress and caregiver strain outcomes.

The third part of Figure 1 is outcomes, represented by T2 measures of the four types of caregiver strain at the end of the 1-year study period. The effects of ComputerLink were evaluated by estimating changes in strain from T1 to T2. This was achieved by controlling for T1 strain in analyses where T2 strain was the dependent variable (Finkel, 1995; Kessler & Greenberg, 1981).

Method

SAMPLE

Caregivers (n = 102) were recruited from a university hospital Alzheimer's registry (n = 40), support groups of a local chapter of the Alzheimer's Association (n = 26), and advertisements about the study (n = 36). Eligibility was limited to primary informal caregivers of community-residing elderly people diagnosed with Alzheimer's disease. Eligible caregivers also had local telephone exchanges and could read and write in English.

Caregivers were randomly assigned to either the ComputerLink (n = 51) or control group (n = 51). Three caregivers in the ComputerLink group were dropped shortly after assignment because of technical problems with installing the computer in their homes. One additional caregiver was not comfortable with the computer and opted out of the study. This left 47 experimental group subjects. Two caregivers from the control group dropped out of the study, leaving 49 and a total sample of 96.

The average age of caregivers was approximately 60 years; 65 (68%) were female and 56 (58%) were spouses, with 22 being husbands and 34 being wives. Of the 42 nonspouse caregivers, 33 (79%) were adult children of the care receiver. About one third of caregivers were employed full- or part-time, and 28% (n = 27) were African American. At the start of the study, caregivers had assisted care receivers for an average of nearly 3 years. Although not a probability sample, study respondents were similar to a nationally representative sample in terms of age, gender, and employment status (Stone, Cafferata, & Sangl, 1987). The sample overrepresented spouses and African Americans.

DATA COLLECTION

All caregivers completed an initial in-person interview (T1) that lasted an average of 90 minutes. A training session for all caregivers in both the experimental and control groups was completed after the T1 interview and randomization. For the experimental group, training focused on the installation and use of ComputerLink. Control group caregivers were provided with an information session on caregiving and Alzheimer's disease. At the end of 12 months, a second in-person interview (T2) was completed, which repeated many of the measures collected in the T1 interview.

In-person interviews followed a structured protocol and had a mix of standardized measures and items developed for this project. Interviewers completed an extensive training session at the outset of the study, and the same interviewer conducted the T1 and T2 survey for each caregiver.

In addition to interview data, ongoing monitoring of all accesses of ComputerLink provided a complete record of use of the system. These data are used to construct measures of the frequency of use of the communication and solitary functions.

MEASURES

Caregiver strain. Caregiver strain was conceptualized as a multidimensional construct that included caregivers' perceptions of physical strain, emotional strain, relationship strain, and activity restriction because of caregiving (Bass, McClendon, Deimling, & Mukherjee, 1994; Deimling, 1994; Deimling & Bass, 1986; Lawton, Kleban, Moss, Rovine, & Glicksman, 1989; Poulshock & Deimling, 1984; Zarit & Teri, 1991). As detailed below, factor analysis with data from this sample confirmed the four expected dimensions of strain. Although each of the four dimensions of strain at T2 were dependent variables, the analysis focuses on ComputerLink's effects on changes in strain from T1 to T2 because measures of T1 strain are entered as predictors of the identical measure of strain at T2 (Finkel, 1995; Kessler & Greenberg, 1981).

Physical strain was a three-item index measuring caregivers' perceptions that caregiving caused them to be sick more often, to have more aches and pains, and to be in worse physical health. In an exploratory factor analysis of all indicators of caregiver strain across the four dimensions, these items formed a separate factor from other strain indicators at both T1 and T2, with loadings from .70 to .90. Cronbach's alphas were .92 at T1 and .90 at T2.

Emotional strain was assessed by three items representing caregivers' perceptions that because of caregiving, they were more nervous or bothered by nerves, felt more irritable, and were more often sad, blue, or downhearted. These items factored separately from items making up the measures of activity restriction and physical strain. However, high cross-loadings were found between items representing emotional strain at T1 and items representing relationship strain at T1. Despite their high correlation at T1, emotional and relationship strain were kept distinct because they factored separately at T2 and were conceptualized as different dimensions. The loadings for emotional strain at T1 and T2 ranged from .54 to .79. Cronbach's alpha was .80 at T1 and .84 at T2.

Relationship strain focused exclusively on the care receiver-caregiver dyed and was composed of five items representing caregiver's perceptions that because of caregiving, they resented their relative, were angry with their relative, felt their relative requested more than was needed, were depressed about their relationship, and felt that their relative tried to manipulate them. One of the five items had a low factor loading of .29 at T1, with the other four ranging from .46 to .84. At T2, loadings were more consistent, with a range of .57 to .81. Cronbach's alpha for relationship strain at T1 was .84 and at T2 was .88.

Activity restriction assessed caregivers' perceptions of decreases in participation in five activities because of caregiving: attending religious services, visiting with friends and family members, participating in group activities, volunteering, and engaging in leisure activities. These five items formed a separate orthogonal factor at T2, with loadings from .52 to .73 and a Cronbach's alpha of .78. The factor structure at T1 was less consistent, with two items having lower than desired loadings. Despite this variability, all five items were retained in constructing the T1 index to be consistent with the T2 measure. Cronbach's alpha for the T1 index was .61.

Although the four dimensions of strain were orthogonal factors at T1 and T2, the unweighted indices were not completely independent of one another. After regressing T2 strain on its T1 measure, the correlations between dimensions ranged from a low of .12 to high of .50. The reader is cautioned that the correlations between dependent variables may increase the chances of finding similar results for different measures of strain (see Kmenta, 1986, pp. 635-648).

ComputerLink use. Three measures of ComputerLink were used. The first, ComputerLink treatment, distinguished between experimental and control group caregivers and was used with the total sample (N = 96). The two other ComputerLink measures were relevant only for the experimental group (n = 47). One represented the average number of times per month that the communication function was used, which included posting or reading messages on the public bulletin board, sending or reading private mail, and posting or responding to messages in the question-and-answer segment. Exploratory factor analysis showed that these three activities formed a distinct orthogonal factor, with loadings from .73 to .89 and a Cronbach's alpha of .84. The other ComputerLink measure represented the average number of accesses per month of selected solitary components of the system, including two modules of the electronic encyclopedia (i.e., one on general information on Alzheimer' disease and the other on self-care for caregivers) and the decision-support module. These three aspects formed a separate factor, with loadings ranging from .77 to .81 and a Cronbach's alpha of .79.

Risk factors for care-related stress. In addition to the strain measures already described, three other variables represented characteristics that make caregivers vulnerable to adverse caregiving consequences: whether the caregiver was the spouse of the care receiver, a lack of informal support at T1, and being the only person living with the care receiver at T1. Variables representing spouse caregivers and living only with the care receiver were dichotomous. The variable for lack of informal support indicated the number of close family members and friends listed by caregivers in their support networks. The scoring of this measure was reversed so that it was consistent with the scoring of the other risk factors, with a higher value representing fewer informal supports.

ANALYSIS

Ordinary least squares regression was used to estimate the buffering effects of ComputerLink, with a separate regression equation estimated for each of the four dimensions of strain at T2. The analysis was in two parts. The first assessed differences in strain over the 12-month study between the experimental and control group. The second examined only the experimental group and assessed whether the use of the two ComputerLink functions was associated with reduced caregiver strain. For both parts of the analysis, product terms were used to test for the interaction of variables measuring care-related stress and the ComputerLink intervention (see Figure 1) (McClendon, 1994). Product terms were created by multiplying each measure of care-related stress by each measure of ComputerLink. Product terms with probabilities less than or equal to .10 were retained in the final equations, whereas nonsignificant ones were dropped using a backward-selection, stepwise procedure. The .10 significance level was used becau e of the small sample size.

Results

Table 1 presents the scoring, means and standard deviations for all variables for the total sample and the experimental group. There were no significant differences between T1 measures for the experimental and control groups, confirming the expected effect of random assignment.

Figures in Table 1 show that slightly more than half of caregivers were husbands or wives of care receivers (58%). Of the nonspouse caregivers, 80% were adult children or daughters-in-law. About two thirds (62%) had no one else in the household except the care receiver. In the experimental group, the communication function was more often used (11.28 per month) than the solitary function (.97 per month). This difference was expected because seeking information is a discrete, episodic event when compared to communication, which includes ongoing conversations.

INTERACTIONS FOR THE EXPERIMENTAL-CONTROL GROUP COMPARISON

The overall effects of the experimental ComputerLink treatment are shown in Tables 2 and 3. Table 2 has one multiple regression equation for each of the four types of caregiver strain at T2. Each equation also contains the T1 measure of strain as an independent variable. Thus, the coefficient for each independent variable, other than the lagged measure of strain, indicates the effect of the variable on change in strain from T1 to T2 (Finkel, 1995; Kessler & Greenberg, 1981). Coefficients of primary interest for the buffering model were the interactions of measures of T1 care-related stress and ComputerLink treatment. Significant interactions in Table 2 would mean that the effect of being in the experimental group and having access to ComputerLink depends on the level of a T1 care-related stress measure. According to the buffering model, the effect of ComputerLink on reducing caregiver strain should be greater for caregivers who are under more stress, or at greater risk of stress, based on their T1 charac eristics (Allison, 1978; Wheaton, 1985). Because a higher score on all of the T1 care-related stress measures indicated greater stress or risk of stress, negative coefficients for the interactions demonstrate a buffering effect.

Table 3 helps to illustrate the multivariate results in Table 2 by presenting conditional mean changes for the experimental and control groups. These means are the average change in strain from T1 to T2 for experimental and control group caregivers with varying scores on the indicators of T1 care-related stress. Although the mean change is used for illustration in Table 3, it should be reiterated that actual T2 scores, not change or difference scores, were the dependent variables in the regression equations. Table 3 also includes the difference between the mean changes for the experimental and control groups (i.e., the experimental mean change minus the control mean change). Negative differences indicate strain decreased more in the experimental group, or the experimental group had a smaller increase in strain than controls. Positive differences indicate that strain decreased more in the control group, or the control group had a smaller increase in strain than the experimental group.

There were no significant product terms for T1 strain and ComputerLink treatment. Thus, participation in the ComputerLink group did not alter the relationship between T1 and T2 caregiver strain.

The variable distinguishing spouse and nonspouse caregivers significantly interacted with ComputerLink treatment to affect relationship strain at T2. The coefficient for the product term was negative (b = -1.58), indicating that the ComputerLink treatment led to a significantly greater reduction in relationship strain for spouses. This effect was consistent with the buffering model, because spouses were considered to be more vulnerable and were expected to benefit more from ComputerLink. The conditional means in Table 3 illustrate the interaction by showing that spouses in the experimental group had a decrease in relationship strain (-.78), whereas control group spouses showed no change (.00). In contrast, nonspouse caregivers in the experimental group showed a small increase (.30) in relationship strain, whereas the control group showed a decrease (-.85).

Lack of informal support significantly interacted with the ComputerLink treatment to affect emotional strain (b = .76) and activity restriction (b = .74). That is, ComputerLink access led to significantly greater reductions in emotional skein and activity restriction for caregivers with more informal support. These interactions were contrary to the buffering model because the benefits of ComputerLink were expected to be greater for vulnerable caregivers who had less informal support. The two interactions are illustrated in Table 3. Among caregivers with above-average informal support, the experimental group showed a greater reduction in emotional strain than the control group, whereas among those with less informal support, the control group had a greater reduction in emotional strain. The difference of -.52 for those with more support and .97 for those with less support summarizes this pattern. For activity restriction, there was a similar pattern of conditional means.

The variable representing the living arrangement of the caregiver (i.e., lives with the care receiver only) significantly interacted with ComputerLink treatment to affect activity restriction (b = 1.92). As illustrated in Table 3, the decrease in activity restriction in the experimental group was greater for caregivers not living alone with care receivers. Similar to the interactions with informal support, this was contrary to the buffering model because it indicated that less vulnerable ComputerLink caregivers, who lived with other relatives, had a greater decrease in activity restriction.

INTERACTIONS FOR THE COMMUNICATION AND SOLITARY FUNCTIONS

Tables 4 and 5 present the effects of using the communication and solitary ComputerLink functions and focus on whether the frequency of use of the communication and solitary functions interacted with T1 care-related stress to influence T2 caregiver strain. Tables 4 and 5 are similar in structure to Tables 2 and 3 with two exceptions. In Table 5, the variables for communication and solitary function were dichotomized to calculate the conditional mean changes. In addition, in Table 5, interactions involving T1 strain used the actual means of the dependent variables rather than the mean change from T1 to T2.

The first row of interactions in Table 4 shows that three T1 strain measures significantly interacted with communication function use: physical strain (b = -.04), relationship strain (b =-.02), and activity restriction (b = -.05). The negative coefficients for these product terms are consistent with the buffering model because they show that caregivers with higher strain at T1, who more frequently used the communication function, had a greater reduction in strain at T2.

For example, Table 5 shows that caregivers with high scores on both T1 physical strain and communication function use had average T2 physical strain scores of 7.25, compared to 8.67 for caregivers high in T1 physical strain but low in communication function use. There was a similar pattern in Table 5 for T2 relationship strain. For T2 activity restriction, the pattern of conditional means in Table 5 does not demonstrate the significant interaction between T1 activity restriction and communication function use. This may result because T1 activity restriction and communication function use are dichotomized into high and low groups to calculate the conditional means or because no other variables are controlled when the means are calculated.

The variables spouse caregiver and communication function use interacted to produce significant effects on T2 physical strain (b = . 13), T2 relationship strain (b = .22), and T2 activity restriction (b = .18). These three interactions, however, are contrary to the buffering model because they show that less vulnerable nonspouse caregivers who more frequently used the communication function had greater reductions in T2 strain than more vulnerable spouse caregivers. For example, the difference scores in Table 5 for T2 physical strain show that nonspouses with high compared to low communication function use differed by -1.48, whereas the difference for spouses with high compared to low communication use was -.44. Similar patterns are illustrated in Table 5 for T2 relationship strain and T2 activity restriction.

The opposite pattern is presented by the significant interaction in Table 4 between the spouse variable and solitary function use when T2 activity restriction is the dependent variable (b = -2.52). This effect was consistent with the buffering model's proposition that more vulnerable spouses would experience greater ComputerLink benefits, as illustrated in Table 5.

The variable lack of informal support significantly interacted with solitary function use (Table 4) for the equation for T2 activity restriction (b = .76). Contrary to expectations, caregivers with more informal support who were high on solitary function use had the greater decrease in activity restriction (Table 5).

The results in Table 4 for the variable lives only with the care receiver show that it interacted with communication function use in the equation for T2 emotional strain and with solitary function use in the equations for all four types of strain. The consistent interactions with solitary function use (T2 physical strain b = -2.81; T2 emotional strain b = -2.12; T2 relationship strain b = -2.39; and T2 activity restriction b = -3.78) were in accordance with the buffering model. Greater reductions in strain were experienced by more vulnerable caregivers who lived alone with care receivers. The conditional means in Table 5, as well as their difference scores, illustrate each of these effects. The interaction between living only with the care receiver and communication function use was positive (b = .22) but indicated that greater use of this function is associated with reduced emotional strain for caregivers who did not live only with the care receiver, contrary to the buffering model (see Table 5).

Discussion

Two research questions were addressed by this study: (a) whether access to ComputerLink reduced several types of caregiver strain and (b) whether the impact of ComputerLink varied by the frequency of use of the communication and solitary functions. The study was guided by the widely used buffering model, which hypothesized that vulnerable caregivers were more likely to benefit from this supportive intervention. This hypothesis implies that the effects of ComputerLink will not be uniform for all caregivers but will depend on initial vulnerability. In summarizing study findings, general trends are highlighted with less attention given to differences across the four types of caregiver strain. This approach enhances the interpretations and usefulness of findings.

The first research question was addressed by the experimental versus control group comparison, with results suggesting that ComputerLink led to reduced levels of strain for some caregivers. Reductions in strain were greater for caregivers who had more informal support and for spouse caregivers. The second research question was addressed by examining patterns of computer use within the experimental group. Results indicated that there were important differences in the effects of the communication and solitary functions. However, reductions in strain associated with the use of these functions did not apply to all caregivers. In general, the use of the communication function was related to reduced strain for caregivers who reported more strain at the start of the study and for nonspouse caregivers, who were mainly adult children. Use of the solitary function was related to reduced strain for caregivers living alone with care receivers and for spouse caregivers.

All of these findings confirmed the buffering model's assumption that the effects of a supportive intervention will differ depending on other characteristics of caregivers. Thus, the model was helpful in selecting variables and designing statistical procedures that represented conditional effects. However, the model's assumption that more vulnerable caregivers would experience greater benefits was not consistently supported. Findings that followed the hypothesized pattern showed greater reductions in strain for spouse caregivers, caregivers living alone, and caregivers with higher initial strain. In contrast, findings that were opposite the hypothesized pattern showed greater reductions in strain for caregivers with more informal support and nonspouse caregivers.

One specific area for additional study relates to findings suggesting that ComputerLink had more beneficial effects for caregivers with more informal support. Although this research only can speculate on the exact explanation for this effect, current findings caution that a computer support network may be most appropriate for caregivers who have a stronger network of family and friends and may be an ineffective intervention for caregivers with weaker support networks. Future research should consider a number of possible explanations for these findings, including the following: Informal helpers provide an opportunity for time away from caregiving, when the computer can be conveniently used; topic areas addressed by the network, such as encouraging caregivers to seek help from family members and friends, are most relevant and helpful for caregivers who have an informal network to call on; and a computer network is most effective as a supplemental source of support but cannot substitute for a failed informa system.

Another important area for further study extends from the different effects of the communication and solitary components of ComputerLink. If replicated, these differences can be used to design computer networks that are more appropriate for meeting the needs of certain caregivers; all functions of a network may not be necessary to achieve a desired outcome. For example, use of the communication function was particularly salient for caregivers who reported elevated levels of strain when the study began. This suggests that communication may have the greatest benefits for the subset of caregivers who are already experiencing care-related difficulties. On the other hand, findings for the solitary function suggest that computer-based information can be helpful regardless of whether caregivers report being stressed, particularly if they are living alone with the care recipient. Thus, information may have the potential of preventing problems for certain caregivers.

Results also suggest that a network established for adult children and other nonspouse caregivers should emphasize the communication aspect because it was related to greater reductions in strain for this group. Further study is needed to understand the underlying causes of this finding with possible explanations, including (a) generational differences may make communication with other caregivers a more helpful source of assistance for younger nonspouse caregivers, or (b) the types of problems faced by nonspouse caregivers may be more amenable to suggestions given by other caregivers via the computer.

Although communication appeared more efficacious for nonspouses, a noteworthy finding was that spouse caregivers had greater reductions in relationship strain in the experimental versus control group comparison. Alzheimer's disease often has devastating effects on the quality of the marital relationship. The strain that this disease places on the relationship alters long-term patterns of emotional and physical expression, increases the risk of abusive behaviors by caregivers and care receivers, and may necessitate a complete reorganization of family and friendship roles (Anetzberger, in press; Miller et al., 1991; Pruchno & Potashnik, 1989; Zarit, 1980). If a computer support network is replicated as a means for softening the negative effects on the marital relationship, this alone would be sufficient justification for recommending broad implementation of this intervention for spouse caregivers.

Because it is one of the first systematic evaluations of the impact of a computer support network (Alemi & Stephens, 1996; Brennan, 1995), this study should be considered exploratory, with results providing direction for future research with larger and more diverse samples. It also should be recognized that many variables were not considered in this analysis, such as the content of information disseminated via the network and the composition of network users. Furthermore, change in caregiver strain was the only outcome examined. Other potential benefits from a computer support network, such as improved knowledge of caregiving, delayed institutionalization, increased use of needed services, and positive caregiving consequences, were not considered.

Table 1 Scoring, Means, and Standard Deviations for All Variables
Legend for Table:
A - Total Sample (N = 96)
B - Experimental Group (n = 47) A B M SD M SD
T1 Care-related stress Physical strain (0-9)[a] 2.91 2.34 3.13 2.32 Emotional strain (0-9)[a] 5.00 2.16 5.02 2.61 Relationship strain (0-12)[a] 2.97 2.74 3.11 2.80 Activity restriction (0-10)[a] 7.62 1.81 7.45 1.67 Spouse caregiver (1 = yes) .58 .50 .57 .50 Lack of informal support 3.41 1.47 3.62 1.41 (higher score = fewer helpers) Lives with care recipient only .62 .49 .70 .46 (1 = yes)
Intervention ComputerLink treament (1 = yes) .49 .50 - - Communication function use - - 11.28 13.35 (average accesses/month) Solitary function use - - .97 1.16 (average accesses/month)
T2 Caregiver strain Physical strain
0-9)[a] 3.51 2.63 3.51 2.65 Emotional strain (0-9)[a] 4.53 2.35 4.60 2.78 Relationship strain (0- 12)[a] 2.64 2.74 2.79 3.00 Activity restriction (0- 10)[a] 6.73 2.49 6.62 2.68
[a.] Higher scores indicate greater strain.

Table 2 Multiple Regression of T2 Caregiver Strain on ComputerLink Intervention and T1 CareRelated Stressors (N = 96)
 Dependent Variables T2 T2 Physical Strain Emotional Strain B Beta B Beta
T1 Care-related stress T1 Strain[a] .60[***] .54 .50[***] .46 Spouse caregiver -.72 -.14 -.27 -.06 Lack of informal -.13 -.07 -.18 -.11 support[a] Lives with care .49 .09 .23 .05 recipient only
Intervention ComputerLink -.29 -.06 -.02 -.00 treatment
Interactions with ComputerLink treatment T1 Strain[a] - - - - Spouse caregiver - - -1.58[*] -.26 Lack of informa - - .76[**] .32 support[a] Lives with care - - - - recipient only Total adjusted R[sup 2] .2
[***] .25[***] T2 T2 Relationship Activity Strain Restriction B Beta B Beta
T1 Care-related stress T1 Strain[a] .68[***] .68 .33[**] .24 Spouse caregiver .59 .11 .78 .15 Lack of informal .35[**] .19 -.43[*] .26 support[a] Lives with care .37 .07 -1.13 .22 recipient only
Intervention ComputerLink .83 .15 -1.25 -.25 treatment
Interactions with ComputerLink treatment T1 Strain[a] - - - - Spouse caregiver Lack of informa - - .74[**] .29 support[a] Lives with care - - 1.92[*] .37 recipient only Total adjusted R[sup 2] .46[***] .09[**]
[a.] The means
for these variables are centered to zero.
[*] p </= .1.
[**] p </= .05.
[***] p </= .01.

Table 3 Mean Change in Strain (T2 - T1) by Experimental-Control Groups and Level of Care-Related Stress for Significant Interactions (N = 96)
Legend for Table:
A - Experimental Group
B - Control Group
C - Experimental Minus Control A B C
Emotional strain/formal support More -.96 -.44 -.52 Less .17 -.80 .97
Relationship strain/spouse caregiver No .30 -.85 1.15 Yes -.78 .00 -.78
Activity restriction/informal support More -1.21 -.56 -.65 Less -.26 -2.40 2.14
Activity restriction/lives with care recipient only No -1.71 -.55 -1.16 Yes -.45 -1.26 .81

Table 4 Multiple Regression of T2 Caregiver Strain on Two ComputerLink Functions and T1 Care-Related Stress for the Experimental Group (n = 47)
 Dependent Variables T2 T2 Physical Strain Emotional Strain B Beta B Beta
T1 Care-related stress T1 Strain[a] .44[**] .38 .66[***] .63 Spouse -.76 -.14 -1.12 -.21 caregiver Lack of informal .06 .03 .62[**] .32 support[a] Lives with care -.28 -.05 -.12 -.02 recipient only
ComputerLink intervention Communication -.05 -.24 -.23[**] -1.11 function use[a] Solitary 2.64[**] 1.16 1.91[*] .81 function use[a]
ComputerLink interactions Communication function use[a] T1 Strain[a] -.04[***] -.56 Spouse .13[*] .54 caregiver Lack of informal support Lives with care .22[***] .98
recipient only Solitary function use[a] T1 Strain[a] Spouse caregiver Lack of informal support[a] Lives with care 2.81[**] -1.16 -2.12[*] -.86 recipient only R[sup 2] change .24[*] .17
Total adjusted R[sup 2] .30[***] .39[***] T2 T2 Relationship Activity Strain Restriction B Beta B Beta
T1 Care-related stress T1 Strain[a] .67[***] .63 .13 .08 Spouse -.71 -.12 .77 .14 caregiver Lack of informal .50[**] .24 .62[**] .32 support[a] Lives with care -.31 -.05 -.22 -.04 recipient only
ComputerLink intervention Communication -.14[***] -.64 -.09 -.45 function use[a] Solitary 2.04
**] .79 3.83[***] 1.67 function use[a]
ComputerLink interactions Communication function use[a] T1 Strain[a] -.02[**] -.26 -.05[**] -.32 Spouse .22[***] .83 .18[***] .75 caregiver Lack of informal support Lives with care recipient only Solitary function use[a] T1 Strain[a] Spouse caregiver -2.52[**] -.81 Lack of informal .76[**] .55 support[a] Lives with care -2.39[**] -.88 3.78[***] -1.56 recipient only R[sup 2] change .17[*] .33[**]
Total adjusted R[sup 2] .60[***] .35[***]
[a.] The means for these variables are centered to zero.
[*] p </= .1.
[**] p </= .05.
[***] p </= .01.

Table 5 Mean Change in Strain (T2 - T1) by Communication and Solitary Function Use and Level of T1 Care-Related Stress (n = 47)
 Communication Function Use High Low High - Low
Physical strain T2 T1 Physical strain Low 6.75 5.27 1.48 High 7.25 8.67 -1.42 Spouse caregiver No .14 1.62 -1.48 Yes -.44 .00 -.44 Lives with care recipient only No Yes
Emotional strain T2 Lives with care recipient only No -3.00 .16 -3.16 Yes -.11 -.21 .10
Relationship strain T2 T1 Relationship strain[a] Low 7.88 7.56 .32 High 9.50 10.62 -1.12 Spouse caregiver No -1.14 1.08 -2.22 Yes .00 1.67 1.67 Lives with care recipient only No Yes
Activity restriction T2 T1 Activity restriction[a]
Low 11.80 10.54 1.26 High 13.17 11.78 1.39 Spouse caregiver No -1.00 -1.62 .62 Yes 1.22 -1.22 2.44 Informal support More Less Lives with care recipient only No Yes Solitary Function Use High Low High - Low
Physical strain T2 T1 Physical strain Low High Spouse caregiver No Yes Lives with care recipient only No -.33 .00 -.33 Yes -.22 .92 -1.14
Emotional strain T2 Lives with care recipient only No -1.60 -1.29 -.31 Yes -1.11 .17 -1.28
Relationship strain T2 T1 Relationship strain[a] Low High Spouse caregiver No Yes Lives with care recipient only No .00 -.88 .88
 Yes -.33 -.21 -.12
Activity restriction T2 T1 Activity restriction[a] Low High Spouse caregiver No -1.80 -1.27 -.53 Yes -1.70 .35 -2.05 Informal support More -2.89 -.42 -2.47 Less .00 .38 .38 Lives with care recipient only No -1.50 -1.88 .38 Yes -1.89 .08 -1.97
[a.] Mean strain at T2 rather than mean change in strain
(T2 - T1).
DIAGRAM: Figure 1. Conceptual model for the buffering effects of ComputerLink.

REFERENCES
Alemi, F., Mosavel, M., Stephens, R. C., Ghadiri, A., Krishnaswamy, J., & Thakkar, H. (1996). Electronic self-help and support groups. Medical Care, 34, OS32-OS44.

Alemi, F., & Stephens, R. C. (1996). Computer services for patients: Description of systems and summary of findings. Medical Care, 34, OS1-OS9.

Allison, P. D. (1978). Testing for interaction in multiple regression. American Journal of Sociology, 83, 144-153.

Anetzberger, G. J. (in press). The ethics of personal safety. In T. Johnson (Ed.), Handbook on ethical issues in aging. Westport, CT: Greenwood.

Bass, D. M., McClendon, M. J., Deimling, G. T., & Mukherjee, S. (1994). The influence of a diagnosed mental impairment on family caregiver strain. Journal of Gerontology, 49, S146-S155.

Bass, D. M., Noelker, L. S., & Rechlin, L. R. (1996). The moderating influence of service use on negative caregiving consequences. Journal of Gerontology, 51B, S121-S131.

Bourgeois, M. S., Schulz, R., & Burgio, L. (1995). Improving outcomes by monitoring the intervention process [Special issue 1]. The Gerontologist, 35, 189.

Brennan, P. F. (1995). Characterizing the use of health care services delivered via computer networks. Journal of the American Medical Informatic Association, 2, 160-168.

Brennan, P. F., Moore, S. M., & Smyth, K. S. (1995). The effects of a special computer network on caregivers of persons with Alzheimer's disease. Nursing Research, 44, 166-172.

Brown-Ewing, L. J., Finkelstein, S. M., Budd, J. R., Kujawa, S. J., Wielinski, C. L., Warwick, W. J., & Nguyen, S. (1988). Implementation of a home-based program for early detection of clinical deterioration in cystic fibrosis. Medical Instrumentation, 22, 240-246.

Cantor, M. H. (1983). Strain among caregivers: A study of experience in the United States. The Gerontologist, 23, 597-604.

Christ, G., & Siegel, K. (1990). Monitoring quality-of-life needs in cancer patients. Cancer, 65, 760-765.

Cicirelli, V. G. (1983). A comparison of helping behavior to elderly parents of adult children with intact and disrupted marriages. The Gerontologist, 23, 619-625.

Clipp, E. C., & George, L. K. (1990). Caregiver needs and patterns of social support. Journal of Gerontology, 45, S102-S111.

Deimling, G. T. (1994). Caregiver functioning. In M. P.Lawton & J. Teresi (Eds.), The annual review of gerontology and geriatrics: Vol. 14. Focus on assessment techniques (pp. 257-280). New York: Springer.

Deimling, G. T., & Bass, D. M. (1986). Symptoms of mental imp impairment among elderly adults and their effects on family caregivers. Journal of Gerontology, 41, 778-784.

Fernsler, J. I., & Manchester, L. J. (1997). Evaluation of a computer-based cancer support network. Cancer Practice, 14, 172-175, 182.

Finkel, S. (1995). Causal analysis with panel data. Thousand Oaks, CA: Sage.

Finney, J. W., Mitchell, R. E., Cronkite, R. C., & Moos, R. W. (1984). Methodological issues in estimating main and interactive effects: Examples from coping/social support and stress field. Journal of Health and Social Behavior, 25, 85-98.

George, L. K., & Gwyther, L. (1986). Caregiver well-being: A multidimensional examination of family caregivers of demented adults. The Gerontologist, 26, 253-259.

Greenberger, M., & Puffer, J. C. (1989). Telemedicine: Toward better health care for the elderly. Journal of Communications, 39, 137-144.

Gustafson, D. H., Hawkins, R. P., Boberg, E. W., Bricker, E., Pingree, S.. & Chan, C. L. (1994). The use and impact of a computer-based support system for people living with AIDS and HIV infection. Medical Care, 32, 604-608.

Gustafson, D. H., Taylor, J. O., Thompson, S., & Chesney, P.(1993). Assessing the needs of breast cancer patients and their families. Quality Management in Health Care, 2, 6-17.

Kessler, R. C., & Greenberg, D. F.(1981).Linear panel analysis: Models of quantitative change. New York: Academic Press.

Kmenta, J. (1986). Elements of econometrics (2nd ed.). New York: Collier-Macmillan.

Knight, B. G., Lutzky, S. M., & Macofsky-Urban, F.(1993). A meta-analytic review of interventions for caregiver distress: Recommendations for future research. The Gerontologist, 33, 240-248.

Kosloski, K., & Montgomery, R.J.V. (1995). The impact of respite use on nursing home placement. The Gerontologist, 35, 67-74.

Lawton, M. P., Brody, E. M., & Saperstein, A. R. (1991). Respite for caregivers of Alzheimer patients: Research and practice. New York: Springer.

Lawton, M. P., Kleban, M. H., Moss, M., Rovine, M., & Glicksman, A. (1989). Measuring caregiver appraisal. Journal of Gerontology: Psychological Sciences, 44, P61-P71.

Lin, N. (1986). Modeling the effects of social support. In N. Lin, A. Dean, & W. Ensel (Eds.), Social support, life events, and depression (pp. 173-209). Orlando, FL: Academic Press.

McClendon, M. (1994). Multiple regression and causal analysis. Itasca, IL: F.E. Peacock.

Miller, B., McFall, S., & Montgomery, A. (1991). The impact of elder health, caregiver involvement, and global stress on two dimensions of caregiver burden. Journal of Gerontology, 46, S9-S19.

Montgomery, R.J.V. (1996). Advancing caregiver research: Weighing efficacy and feasibility of interventions. Journal of Gerontology, 51B, S109-S110.

Montgomery, R.J.V., & Borgatta, E. F.(1989). The effects of alternative support strategies on family caregiving. The Gerontologist, 29, 457-464.

Paperny, D. M., & Starn, J. R. (1989). Adolescent pregnancy prevention by health education computer games: Computer-assisted instruction of knowledge and attitudes. Pediatrics, 83, 742-752.

Poulshock S. W., & Deimling, G. T. (1984). Families caring for elders in residence: Issues in the measurement of burden. Journal of Gerontology, 39, 230-239.

Pruchno, R. A., & Potashnik, S. L. (1989). Caregiving spouses: Physical and mental health in perspective. Journal of the American Geriatrics Society, 37, 697-705.

Robinson, T. N. (1989). Community health behavior change through computer network health promotion: Preliminary findings from Stanford Health-Net. Computer Methods Programs Biomed, 30, 137-144.

Schmall, V. L. (1984). It doesn't just happen: What makes a support group good? Generations, 9, 64-67.

Slatalla, M. (1996, May). Who can I turn to? Wired pp. 116-186.

Smyth, K. A., & Harris, P.B. (1993). Using telecomputing to provide information and support to caregivers of persons with dementia. The Gerontologist, 33, 123-127.

Southwood, K. E. (1978). Substantive theory and statistical interaction: Five models. American Journal of Sociology, 83, 1154-1203.

Stone, R., Cafferata, G. L., & Sangl, J. (1987). Caregivers of the frail elderly: A national profile. The Gerontologist, 27, 616-626.

Toseland, R. W., & Rossiter, C. M. (1989). Group interventions to support family caregivers: A review and analysis. The Gerontologist, 29, 438-448.

Wheaton, B. (1985). Models for the stress-buffering functions of coping resources. Journal of Health and Social Behavior, 26, 352-364.

Whidlatch, C. J., Zarit, S. H., & von Eye, A. (1991). Efficacy of interventions with caregivers: A reanalysis. The Gerontologist, 31, 9-13.

Wright, L. K., Clipp, E. C., & George, L. K. (1993). Health consequences of caregiver stress. Medicine, Exercise, Nutrition and Health, 2, 181-195.

Wright, S. D., Lund, D. A., Pett, M. A., & Caserta, M. S. (1987). The assessment of support group experience by caregivers of dementia patients. Clinical Gerontologist, 6, 35-59.

Zarit, S. H. (1980). Aging and mental disorders. New York: Free Press.

Zarit, S. H., & Teri, L. (1991). Interventions and services for family caregivers. Annual Review of Gerontology and Geriatrics, 11, 287-310.

~~~~~~

By DAVID M. BASS, PhD The Benjamin Rose Institute , MCKEE J. MCCLENDON, PhD University of Akron , PATRICIA FLATLEY BRENNAN, RN, PhD, FAAN University of Wisconsin-Madison and CATHERINE MCCARTHY The Benjamin Rose Institute