Exploring the relationship between frequent internet use and health and social care resource use in a community-based cohort of older adults: an observational study in primary care

Objectives Given many countries’ ageing populations, policymakers must consider how to mitigate or reduce health problems associated with old age, within budgetary constraints. Evidence of use of digital technology in delaying the onset of illness and reducing healthcare service use is mixed, with no clear consensus as yet. Our aim was to investigate the relationship between frequent internet use and patterns of health or social care resource use in primary care attendees who took part in a study seeking to improve the health of older adults. Methods Participants recruited from primary care, aged >65 and living in semirural or urban areas in the south of England, were followed up at 3 and 6 months after completing a comprehensive questionnaire with personalised feedback on their health and well-being. We performed logistic regression analyses to investigate relationships between frequent internet use and patterns of service use, controlling for confounding factors, and clustering by general practitioner practice. Four categories of service use data were gathered: use of primary National Health Service (NHS) care; secondary NHS care; other community health and social care services; and assistance with washing, shopping and meals. Results Our results show, in this relatively healthy population, a positive relationship (OR 1.72, 95% CI 1.33 to 2.23) between frequent internet use and use of any other community-based health services (physiotherapist, osteopath/chiropractor, dentist, optician/optometrist, counselling service, smoking cessation service, chiropodist/podiatrist, emergency services, other non-specific health services) and no relationship with the other types of care. No causal relationship can be postulated due to the study’s design. Conclusions No observed relationship between frequent internet use and primary or secondary care use was found, suggesting that older adults without internet access are not disadvantaged regarding healthcare use. Further research should explore how older people use the internet to access healthcare and the impact on health.

relationships. The objective of this study was to investigate the relationship between frequent Internet use and use of different types of services. Literature has suggested that Internet use is an important indicator of socio-economic status (SES), or digital divide mirrors the gaps in SES. The apparent relationship between Internet use and health service use was actually the relationship between SES and health service use. The data showed that frequent Internet use was significantly associated with use of specialized health services (therapist, dentist, etc.) but no relationship with primary or secondary care. The data were consistent with literature, i.e., people with higher SES are more likely to use specialized services; and people of low SES only use basic health services. Your study did not measure SES (education, income, welfare status) and therefore could not address the important confounding effect. For any relationship study, the selection of covariates was determined by the study purpose, and each covariate needs clear rationale for being included in the model. Most covariates were included without any explanation, for example, loneliness scale, season at start. Further, why SF24 was used and why it was broken down to two scores? One of the outcome variables was "wash/meals assistance", an important indicator of disability. SF-12 physical score is a good measure of physical health or disability, so no surprise that it"s a strong predictor of using wash/meals assistance. Therefore to examine the relationship between frequent use and "wash/meals assistance" is less meaningful without explaining the relationship between disability and wash/meals assistance. Other comments: The Introduction should go straight to the purpose of the study and build up the significance and rational of the study. The Introduction has limited literature on digital divide and health disparities but includes lots of information about MRA-O and WISH, it appears the study was driven by the data from WISH not the other way around. The Methods should follow the format of published articles and detail data sources, sampling and collection methods. If scales were used, Cronbach alpha should be reported. The Data Analysis section should describe whether hierarchical modelling was used given that data were collected from multi-stage sampling strategy and longitudinal design. Why GP surgery contribution was used as random effect needs explanation too. How missing data were handled needs to be included in Methods. If only complete cases were included in data analysis, the actual sample size should be reported. Results: the tables should be formatted following examples of published articles; usually % is sufficient when describing sample characteristics. When reporting relationship (either binary or multivariate) it"s a common practice to indicate p-value. Discussion: it"s important to discuss the study within the existing literature, especially related studies on older adult"s digital divide and health disparities.
Please leave your comments for the authors below This is a well written paper looking at internet use and health care service access amongst older adults. However, the paper is limited by the study design, which is acknowledged by the authors. The data is from a larger survey looking at the feasibility of implementing a risk-appraisal system. However, the survey tool failed to ask respondents how they use the internet e.g. for email, travel or health information etc. It also failed to ask respondents how they accessed the internet e.g. computer, i-pad, or smart phone; information that is potentially useful in the development of online information portals and new resources targeting older adults. These oversights severely limit its contribution to our understanding of internet use by older adults.
-We acknowledge these limitations and thank the reviewer for their comments. This paper contributes to the growing evidence that older people are increasingly using the internet and the implications for communications with primary health care services. It was not the aim of our analysis to develop online information portals or new resources, more to understand how access to the internet may influence health outcomes and service use in older people.
The methodology section should be strengthened by inclusion of information on how the participants were identified and recruited to the study.
-We thank the reviewer for this suggestion, and have added information on identification and recruitment to the paper (page 5, Data collection paragraph).
There is a lack of discussion regarding the impact of known demographic variables on internet access. This is important, particularly in relation to the discussion relating to the drive to increase the presence and activity of GP practices online.
-We agree that this is an important discussion point, and have made additions in the Discussion relating to these points (see page 12, last two paragraphs; and page 13, first paragraph). This paper reports a cross-sectional analysis of Internet use and use of health services in older primary care attendees. Given that data on older adults" (aged >65) Internet use were limited, the study has its merit and significance.
-We thank the reviewer for this observation.

Major comments:
The selection of covariates needs to consider possible confounding relationships. The objective of this study was to investigate the relationship between frequent Internet use and use of different types of services. Literature has suggested that Internet use is an important indicator of socio-economic status (SES), or digital divide mirrors the gaps in SES. The apparent relationship between Internet use and health service use was actually the relationship between SES and health service use. The data showed that frequent Internet use was significantly associated with use of specialized health services (therapist, dentist, etc.) but no relationship with primary or secondary care. The data were consistent with literature, i.e., people with higher SES are more likely to use specialized services; and people of low SES only use basic health services. Your study did not measure SES (education, income, welfare status) and therefore could not address the important confounding effect.
-We thank the reviewer for these comments. We used "age at which left full-time education" as a proxy for SES (before 17 years of age = low SES; 17 or older = high SES), as has been done in other published research (1) (2).
-We note that this proxy measure is likely to differ in its appropriateness in the US and UK systems, The effect of SES on health care use may be different in the UK compared to the US due to differences in health system structure. In the UK, there is no difference in NHS health services offered on the basis of SES, as all care, both basic and specialised, is free at the point of use (with some small exceptions, e.g. some prescription charges).
-The group of services that we have found is associated with internet use, in the UK setting, are part private and part public, in the sense that some have co-payments (dentists, opticians) or are private in parallel with public (chiropodists, counsellors and physiotherapists), and some are wholly private (chiropractors) or wholly public (emergency services and smoking cessation services). Use of the internet could be implicated in a person"s ability to find any of these community-based services, except perhaps the emergency services.
-We have added in further clarification and discussion of this point in the Methods section (see page 6-7, Covariates paragraph) and the Discussion section (see page 12, first paragraph of Discussion section).
For any relationship study, the selection of covariates was determined by the study purpose, and each covariate needs clear rationale for being included in the model. Most covariates were included without any explanation, for example, loneliness scale, season at start. Further, why SF24 was used and why it was broken down to two scores? -We thank the reviewer for highlighting this. We have added justification for inclusion of each covariate, and we have explained more clearly that the SF-12 is commonly assessed and reported as its two component scoresmental and physical (see Methods, top of page 7).
One of the outcome variables was "wash/meals assistance", an important indicator of disability. SF-12 physical score is a good measure of physical health or disability, so no surprise that it"s a strong predictor of using wash/meals assistance. Therefore to examine the relationship between frequent use and "wash/meals assistance" is less meaningful without explaining the relationship between disability and wash/meals assistance.
-In the UK context, "wash/meal assistance" is important as an indicator that the person has a disability serious enough to need help with personal care. For many older people their health (physical or mental) may limit the activities listed within the SF-12 (e.g. vacuuming, playing golf, climbing several flights of stairs) but a much smaller group needs personal care related to keeping clean and assistance with food. This variable therefore identifies a smaller more impaired sub-group. We feel it is valuable therefore to explore the relationship between internet use and this type of care.
-We have added a sentence to the discussion to clarify this pointwe would expect this group to have some degree of disability and therefore also potentially less internet use, but we did not demonstrate this in our study (see page 12, second paragraph).
Other comments: The Introduction should go straight to the purpose of the study and build up the significance and rational of the study. The Introduction has limited literature on digital divide and health disparities but includes lots of information about MRA-O and WISH, it appears the study was driven by the data from WISH not the other way around.
-We thank the reviewer for this observation. We have amended the Introduction section and included more detail on the digital divide (see page 4, third and fourth paragraphs).
The Methods should follow the format of published articles and detail data sources, sampling and collection methods. If scales were used, Cronbach alpha should be reported.
-We thank the reviewer for this observation. We have added some extra elements to the Methods section to reflect these helpful suggestions (see page 5).
The Data Analysis section should describe whether hierarchical modelling was used given that data were collected from multi-stage sampling strategy and longitudinal design. Why GP surgery contribution was used as random effect needs explanation too. How missing data were handled needs to be included in Methods. If only complete cases were included in data analysis, the actual sample size should be reported.
-We thank the reviewer for these suggestions; we have incorporated them into the relevant sections and given explanations where required (see page 7, Analysis and Missing data sections, and Table  2).
Results: the tables should be formatted following examples of published articles; usually % is sufficient when describing sample characteristics. When reporting relationship (either binary or multivariate) it"s a common practice to indicate p-value.
-We thank the reviewer for this suggestion, and we have changed the formatting of the tables to reflect these points. Where the denominator is not the same throughout a column/row, we have indicated how many missing values there are.
Discussion: it"s important to discuss the study within the existing literature, especially related studies on older adult"s digital divide and health disparities.
-We agree that this would strengthen the paper and have added further discussion to take these points into account (see page 12, last two paragraphs; and page 13, first two paragraphs).