This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Although the Internet is commonly used to recruit samples in studies of human immunodeficiency virus (HIV)-related risk behaviors, it has not been used to measure patient-reported well-being. As the burden of long-term chronic HIV infection rises, the Internet may offer enormous potential for recruitment to research and interventions.
This study aimed to compare two samples of gay men living with HIV, one recruited via the Web and the other recruited in outpatient settings, in terms of self-reported physical and psychological symptom burden.
The Internet sample was recruited from a UK-wide Web-based survey of gay men with diagnosed HIV. Of these, 154 respondents identified themselves as resident in London and were included in this analysis. The HIV clinic sample was recruited from five HIV outpatient clinics. Of these participants, 400 gay men recruited in London clinics were included in this analysis.
The Web-based sample was younger than the clinic sample (37.3 years, SD 7.0 vs 40.9 years, SD 8.3), more likely to be in paid employment (72.8%, 99/136 vs 60.1%, 227/378), less likely to be on antiretroviral therapy (ART) (58.4%, 90/154 vs 68.0%, 266/391), and had worse mean psychological symptom burden compared to the clinic sample (mean scores: 1.61, SD 1.09 vs 1.36, SD 0.96) but similar physical symptom burden (mean scores: 0.78, SD 0.65 vs 0.70, SD 0.74). In multivariable logistic regression, for the physical symptom burden model, adjusted for age, ethnicity, employment status, and ART use, the recruitment setting (ie, Web-based vs clinic) was not significantly associated with high physical symptom score. The only variable that remained significantly associated with high physical symptom score was employment status, with those in employment being less likely to report being in the upper (worst) physical symptom tertile versus the other two tertiles (adjusted OR 0.41, 95% CI 0.28-0.62,
Our data have revealed a number of differences. Compared to the clinic sample, the Web-based sample had worse psychological symptom burden, younger average age, higher prevalence of employment, and a lower proportion on ART. For future research, we recommend that Web-based data collection should include the demographic variables that we note differed between samples. In addition, we recognize that each recruitment method may bring inherent sampling bias, with clinic populations differing by geographical location and reflecting those accessing regular medical care, and Web-based sampling recruiting those with greater Internet access and identifying survey materials through specific searches and contact with specific websites.
Research protocols that utilize electronic and Web-based methods of participant recruitment to research and intervention participation and associated data collection activities have become increasingly common. The method has become particularly well used in behavioral surveillance research studies among persons living with human immunodeficiency virus (HIV) infection in high-income countries. Web-based methods have been used in various ways in epidemiological HIV studies investigating risk behavior [
A review of the methodological implications of Web-based HIV behavioral surveillance methods summarized the advantages as convenience, reduced costs in the management of tools and data collection and entry, ease of tool modification, anonymity, and reduced social desirability bias [
A review of HIV behavioral research among men who have sex with men (MSM) found equivocal evidence in the literature as to whether men who use the Internet are more likely to report risk behavior, although those who use the Internet for sex are more likely to be younger, report sex with women, have had a sexually transmitted infection, and to use a public sex environment [
A comparison of London MSM recruited via the Web and in the community found that the Web-based sample were younger and were less likely to exclusively have sex with men, to be in a relationship with a man, to have received higher education, or to have been tested for HIV [
There have been methodological advances in the design and implementation of Web-based recruitment and data collection to investigate behavioral aspects of HIV infection (principally primary prevention and risk behavior). However, the utility of using these methods to investigate disease-oriented variables and patient self-report burden of disease has not been explored. As much health research and delivery becomes oriented to long-term and chronic conditions (where the patient-reported experience is an important area of inquiry, and those living with long-term chronic conditions have greater potential to use the Internet as compared to rapidly declining conditions), there may be great potential for use of the Internet for recruitment and data collection for both research and care activities. The rise in the use of patient-reported outcome measures (PROMS) to improve equity and quality in health care [
This study aimed to compare two samples of gay men living with HIV, one recruited via the Web and the other recruited in HIV outpatient settings, in terms of self-report physical and psychological symptom burden. The outcome of interest was the self-report 7-day period prevalence and burden of physical and psychological symptoms.
The study is a secondary analysis of two datasets (one recruited via the Web and one in outpatient clinics). Participants were gay men with diagnosed HIV who were resident in London, United Kingdom.
We summarize the two prior study designs here. The Web-based sample [
The HIV clinic sample [
In both studies, symptoms were measured using the Memorial Symptom Assessment Scale−Short Form (MSAS-SF), a patient self-report scale that measures the 7-day period prevalence of 26 physical and 6 psychological symptoms. This standardized symptom questionnaire captures the presence of each symptom and associated distress (for physical symptoms) or frequency (for psychological symptoms) and has often been reported in studies of people living with HIV [
Respondents in both studies gave demographic data on age (analyzed as a continuous variable), ethnicity (categorized as white/non-white), education (categorized as university/non-university), employment (categorized as currently in paid employment or not), and current antiretroviral therapy (ART) use (yes/no).
For each sample (clinic and Web-based), we present descriptive analyses for the demographic characteristics and MSAS-SF variables (total number of symptoms, global distress subscale, physical distress subscale, and psychological distress subscale). The demographic variables were compared between samples (clinic vs Web-based) using
The comparison of characteristics between the two samples is presented in
The multivariable analysis assessed the effect of setting on symptom score (predicting having a score in the highest tertile compared to the other two) after adjusting for all factors with
Univariate comparison of the clinic versus Web-based samples.
|
Clinic setting (n=400) | Web-based setting (n=154) | Test comparison | Degrees of freedom | ||
|
|
|||||
|
Missing: clinic n=6; Web-based n=1 |
|
||||
|
|
40.9 (SD 8.3) 40.0 | 37.3 (SD 7.0) 37.0 |
|
530 | |
|
|
|||||
|
Missing: clinic n=5; Web-based n=11 |
|
||||
|
|
Below university | 196 (49.6) | 62 (43.1) | χ2=1.65 |
1 |
University | 199 (50.4) | 81 (56.3) | ||||
|
|
|||||
|
Missing: clinic n=6; Web-based n=0 |
|
||||
|
|
White | 346 (87.8) | 143 (92.9) | χ2=2.93 |
1 |
Non-white | 48 (12.2) | 11 (7.1) | ||||
|
|
|||||
|
Missing: clinic n=22; Web-based n=18 |
|
||||
|
|
Not in employment | 151 (39.9) | 37 (27.2) | χ2=7.00 |
1 |
In employment | 227 (60.1) | 99 (72.8) | ||||
|
|
|||||
|
Missing: clinic n=9; Web-based n=0 |
|
||||
|
|
Not on ART | 125 (32.0) | 64 (41.6) | χ2=4.49 |
1 |
On ART | 266 (68.0) | 90 (58.4) | ||||
|
|
|||||
|
Missing: clinic n=0; Web-based n=12 |
|
||||
|
|
1.15 (SD 0.79) 1.12 | 1.25 (SD 0.86) 1.23 |
|
529 | |
|
|
|||||
|
Missing: clinic n=0; Web-based n=17 |
|
||||
|
|
0.78 (SD 0.65) 0.73 | 0.70 (SD 0.74) 0.47 |
|
517 | |
|
|
|||||
|
Missing: clinic n=0; Web-based n=4 |
|
||||
|
|
1.36 (0.96) 1.33 | 1.61 (SD 1.09) 1.65 |
|
522 | |
|
|
|||||
|
<0.72 | 137 (34.3) | 45 (31.7) |
|
|
|
|
0.72-1.52 | 136 (34.0) | 45 (31.7) |
|
|
|
|
>1.52 | 127 (31.8) | 52 (36.6) | χ2(trend)=0.87 |
2 | |
|
|
|||||
|
<0.27 | 131 (32.8) | 55 (40.1) |
|
|
|
|
0.27-1.0 | 126 (31.5) | 47 (34.3) |
|
|
|
|
>1.0 | 143 (35.8) | 35 (25.5) | χ2(trend)=4.66 |
2 | |
|
|
|||||
|
<0.87 | 139 (34.8) | 47 (31.3) |
|
|
|
|
0.87-1.87 | 149 (37.3) | 42 (28.0) |
|
|
|
|
>1.87 | 112 (28.0) | 61 (40.7) | χ2(trend)=4.32 |
2 |
aART: antiretroviral therapy
bMSAS: Memorial Symptom Assessment Scale
Multivariable logistic analysis with (1) physical, and (2) psychological symptoms as independent variable, predicting upper tertile (ie, worst) symptom burden score.
|
Physical symptoms as dependent variable (n=482) | Psychological symptoms as dependent variable (n=495) | |||||
Adjusted odds ratio (95% CI) |
|
Adjusted odds ratio (95% CI) |
|
||||
|
|||||||
|
Clinica | 1 | 1 | ||||
|
Web-based | 0.81 (0.50-1.30) |
|
2.20 (1.41-3.44) |
|
||
|
|
1.00 (0.97-1.02) |
|
0.98 (0.95-1.00) |
|
||
|
|||||||
|
Non-whitea | 1 | 1 | ||||
|
White | 1.40 (0.73-2.69) |
|
1.22 (0.64-2.24) |
|
||
|
|||||||
|
Not employeda | 1 | 1 | ||||
|
Employed | 0.41 (0.28-0.62) |
|
0.32 (0.21-0.49) |
|
||
|
|||||||
|
Noa | 1 | 1 | ||||
|
Yes | 1.39 (0.90-2.15) |
|
1.20 (0.78-1.85) |
|
areference category
bART: antiretroviral therapy
Our data have revealed a number of differences between samples recruited via the Web and in clinic settings, in terms of both demographics and self-report psychological and physical symptom burden. First, as has been found with behavioral studies of gay men, the Web-based sample was younger than the clinic sample. Second, we found that those recruited via the Web were more likely to be in employment, which may reflect the costs associated with Internet connectivity. It may also be that those in employment face greater challenges in attending clinics regularly, or that the clinic sample had a higher proportion of participants with health problems linked to not being employed. Third, the clinic sample was more likely to be on ART. This may be because clinic sampling may under-sample those with poor or erratic attendance who are not on treatment or those with mild immunosuppression not yet on treatment and who feel less need to attend for care. In terms of the self-report symptom burden, participants in the Web-based sample were less likely to have high physical symptom burden; this difference was largely explained by the differences between the samples in demographic characteristics. However, the Web-based sample was more likely to have higher psychological symptom burden; this difference was not attenuated in adjusted analysis. There are several potential explanations for this latter finding. Psychological problems are more prevalent among HIV-infected populations (even among those on ART [
Our findings that the Web-based sample was statistically significantly younger and more likely to be employed are in line with previous studies [
There are a number of limitations to our study. First, although we were able to compare a well-defined population by analyzing data from only those men in the Web-based survey with a London postal code to those accessing care at a London clinic, we note that men may travel to access care. Second, although we had a high response rate for the clinic survey (86%), we do not have a response rate for the Web-based survey.
For future research, we recommend that Web-based recruitment should include collection of the demographic variables that we note differed between samples: age, ART use, and employment. The effect of adjustment for these factors in analyses can then be examined. In addition, we recognize that each recruitment method may bring inherent sampling bias, with clinic populations differing by geographical populations served and reflecting those accessing regular medical care, and Web-based sampling recruiting those with greater Internet access and identifying survey materials through specific searches and contact with specific websites. Furthermore, our Web-based sampling did not allow specification of a sampling frame.
Patient-reported symptom data can feasibly be collected though Web-based recruitment as well as through clinic-based questionnaire studies. There may be some specific advantages of Web-based studies when investigating stigmatized problems such as psychological burden of HIV disease, where social desirability may bias traditional face-to-face recruitment methods. We conclude that the Web-based sample had a higher psychological symptom burden (but not physical symptom burden) and that therefore they differ clinically from the sample recruited in clinics. This offers great opportunity to recruit people with HIV to interventions to improve mental health, which is a highly prevent and burdensome problem in this population.
antiretroviral therapy
human immunodeficiency virus
men who have sex with men
Memorial Symptom Assessment Scale−Short Form
Memorial Symptom Assessment Scale−Physical Distress Subscale
Memorial Symptom Assessment Scale−Psychological Distress Subscale
Memorial Symptom Assessment Scale−Global Distress Subscale
We would like to thank the volunteers of GMFA who conducted the Web-based survey and GlaxoSmithKline for an unrestricted educational grant for the clinic survey.
None declared.