The diffusion of the internet has reduced informational frictions and given consumers unprecedented sources of information about the price and quality of products. A remarkably high proportion of internet users look online for information on health and healthcare products (Fox and Duggan 2013). Access to health information via the web is redefining the roles of ‘supplier’ and ‘consumer’, as the flow of information to patients is no longer controlled by medical practitioners (Hartzband and Groopman 2010).
In principle, having access to broader information should lead to better healthcare decisions. But the merits of more information in healthcare are complex (Phelps 1992). There are concerns about the quality of the information provided being difficult to interpret, about the capacity of users to make use of it (Eysenbach et al. 2002), and about patients without internet access (Wagner et al. 2005). This issue has come to the fore recently with the advent of a coronavirus vaccine and concerns that the internet will lead to ‘anti-vax’ behaviour (Aksoy et al. 2020).
Since broadband internet has diffused tremendously in the past decades, suitable data offer researchers the opportunity to assess the impact of internet access on healthcare behaviour. But establishing this is difficult due to potential endogeneity of internet diffusion. Internet subscription is positively correlated with several observable demographic characteristics (such as income and education) that are also positively correlated with healthcare use and health outcomes. In a recent paper (Amaral-Garcia et al. forthcoming), we address this issue in the context of the demand for Caesarean section births (henceforth, ‘C-sections’). Such deliveries are rising worldwide, despite being more expensive and often unnecessary on medical grounds for many mothers and babies.
We utilise an identification strategy based on exogenous discontinuities in internet quality and access in the UK in the 2000s. The discontinuities we exploit stem from two key elements of the asymmetric digital subscriber line (ADSL) technology, which was by far the most important way to gain access to the internet in these initial years of UK internet rollout. The first is the decay of the digital signal, which means that the quality of the internet connection strongly depends on the distance between the starting node of the connection (the local exchange, or LE) and the delivery point (the house). The second is that the areas served by LEs are irregularly shaped because the topology of the network was designed in the 1930s to serve a different purpose (telephone analogue voice communications).
We use a detailed map of the topology of the internet network to identify adjacent small local areas of around 600 households that are served by different LEs. Because LE catchment areas are irregularly shaped and vary in size, contiguous small local areas can be located at very different distances from their respective node of the network and so will experience (potentially large) differences in quality of internet access. By matching data on LE coverage to detailed census and hospital discharge data, we identify a subset of adjacent small areas that are balanced both in terms of aggregate demographics and patient characteristics but differ with respect to the quality and the availability of internet access (see Figure 1). This identification strategy is possible thanks to the relatively slow (compared to other countries, and notably to the US) development of UK’s broadband internet infrastructure which, in the years that we consider, had an average broadband speed of 1-2Mbit/s, mostly based on the upgrade of the pre-existing telephone network.
Figure 1
Notes: Left panel: The map shows the borders of LEs (bold dashed lines), the locations of the LEs (black dots with LEs’ names), and the LSOAs areas (light blue areas delimited by light grey lines). An example of matched LSOAs is given by the pair of LSOAs filled in red and green. They share a border but belong to two different LEs. The LSOA below the border (in red) is connected to a close LE labelled LWWIL while the LSOA above the border (in green) is connected to a far LE labelled LWCRI. Right panel: average predicted real internet speed over time, in close and far LSOA, weighted by overall internet penetration.
Source: Amaral-Garcia et al. (forthcoming).
The case of childbirth is relevant for several reasons. First, as childbirth is one of the most commonly performed medical procedures worldwide, C-sections are one of the most common surgical procedures. The number of these cases not only gives us the statistical power to perform our analysis, but it also means that these cases are quantitatively relevant, particularly as there is widespread concern over the rise in C-sections which has occurred in many healthcare markets. Second, in the UK the vast majority of mothers give birth in hospitals, which eliminates concerns of a possible selection bias. Third, mothers know their pregnancy status well in advance, giving them time to search for information should they wish to do so. Fourth, childbirth in the UK is tax-financed and free at point of delivery, so all consumers face the same zero price. Fifth, medical staff in the NHS are salaried and do not have financial incentives to undertake higher volumes and more expensive procedures. This feature of the NHS – coupled with the facts that the hospitals used by mothers in the adjacent small area in our analysis are almost always the same and that in those hospitals, the mothers will be treated by the same staff – means we can shut off supply-side drivers of variation in delivery methods.
We find that mothers with better, faster access to the internet are 2.5% more likely to have a C-section than mothers living in areas with worse internet access (Figure 2). This effect comes from an increase in elective C-sections; we find no effect of the internet on the likelihood of performing an emergency C-section (a decision made by the medical supplier and not the demander). We find no effect on either mothers’ or newborns’ health outcomes. Finally, the increase is driven by first-time and low-income mothers, who are 6% more likely to have a C-section. By the end of the period, the C-section ‘gap’ between high- and low-income mothers that is apparent pre-broadband internet is virtually closed, thanks in large part to the internet.
Figure 2
Notes: C-section rates over time. Close LSOAs are reported in black (solid line and circles) while far LSOAs are reported in light grey (dashed line and diamonds).
Source: Amaral-Garcia et al. (forthcoming).
Overall, the internet was a non-negligible factor contributing to the observed increase in C-sections in the first decade of 2000s, when the C-section rate increased from 20.7% in 2000 to 24.3% in 2011 (a total increase of 17.4%). A back-of-the-envelope calculation of the financial cost associated with the increase in C-sections due to internet diffusion during the 2000s was, at a minimum, around £6.5 million per year, with no corresponding medical benefits.
Our findings support a mechanism of information gathering and joint decisions in a setting where the quality of information is heterogeneous. The internet is a very diverse source of information and one much less controlled by experts than information previously available in print and TV media. Hence, mothers are exposed to a wider range of signals online. First-time mothers have less experience of and knowledge about the pros and cons of different delivery methods, and thus they are more likely to search for, and to be influenced by, online information. In line with this mechanism, we find that the increase in C-sections is driven by first-time mothers who opt more often for an elective C-section, while we do not find a difference for multiple-time mothers or for emergency C-sections. We also show that the effect we find is not driven by mothers selecting hospitals with high C-section rates or traveling longer distances to find a hospital to have a C-section. This supports the idea that access to information gives mothers the ability to influence the decision of their normal supplier.
References
Aksoy, C G, B Eichengreen and O Saka (2020), “Vaccine challenges”, VoxEU.org, 16 November.
Amaral-Garcia, S, M Nardotto, C Propper and T Valletti (forthcoming), “Mums Go Online: Is the Internet Changing the Demand for Healthcare?”, Review of Economics and Statistics (also CEPR Discussion Paper 13625).
Burki, T (2019), “Vaccine misinformation and social media”, The Lancet Digital Health 1(6): E258-9.
Eysenbach, G, J Powell, O Kuss and E R Sa (2002), “Empirical Studies Assessing the Quality of Health Information for Consumers on the World Wide Web: A Systematic Review”.
Fox, S and M Duggan (2013), “Health online 2013”, Pew Internet & American Life Project.
Hartzband, P and J Groopman (2010), “Untangling the web - patients, doctors, and the internet”, New England Journal of Medicine 362(12): 1063–1066.
Phelps, C E (1992), “Diffusion of Information in Medical Care”, The Journal of Economic Perspectives 6(3): 23–42.
Wagner, T H, M K Bundorf, S J Singer and L C Baker (2005), “Free Internet Access, the Digital Divide, and Health Information”, Medical Care 4: 415-20.
