Since recruitment through the Internet is less intensive compared with traditional methods, an advantage of the online approach is that cohorts can recruit for many years. Accordingly, NINFEA is a dynamic cohort with ongoing recruitment and a minimum target of 7,500 participants; however, we report data last downloaded in March 2015. The ELF cohort was a feasibility study and obtained a minimum target of 5,000 participants; recruitment ended at the end of 2012.

The NINFEA cohort started as a pilot study in the city of Turin, Italy, in July 2005 and has been gradually extended to the rest of Italy. The original study protocol and subsequent amendments have been approved by the Ethical Committee of the San Giovanni Battista Hospital—CTO/CRF/Maria Adelaide Hospital, Turin, Italy. Members of the cohort are children of mothers who have access to the Internet, have enough knowledge of the Italian language to complete an online questionnaire, and volunteer to participate at any time during the pregnancy. They register through the project Web site and complete the first questionnaire that lasts approximately 30 minutes. While the Web site has always been public and accessible from any part of Italy (and the world), the methods for advertisement of the existence of the study have changed over time.

All women participate online, although NINFEA is advertised using both offline and online methods. Offline methods involve the collaboration of health personnel and, therefore, target a prespecified catchment population. Currently, the NINFEA study is actively advertised in the city of Turin, in the Tuscany Region and, with a lower intensity, in the Piedmont Region (of which, the city of Turin is the capital). In these areas, leaflets and posters were distributed, and the study was introduced to pregnant women when they attended hospitals or family clinics for reasons related to their pregnancy. Online recruitment includes recruitment through the Internet (eg, Web sites, forums, social networks) and the media.Until March 2015, approximately 16% of participants were recruited via a passive mode, 82% were recruited actively, and 2% of participants comprised both modes. A total of 7003 pregnant women were recruited in the study as of March 2015. The 3 most represented Italian Regions in the NINFEA cohort are the Piedmont Region (62% of the participants), the Tuscany Region (22%), and the Lombardy Region (4%), while the most represented municipality is Torino (45%). About one-third of the participants are from central urban areas, almost 50% from peripheral urban areas, and the remaining 20% are from rural areas.

The primary location for the ELF cohort is Wellington, New Zealand, but additional study sites are located in the other main city centers (eg, Christchurch and Auckland). Ethical approval was obtained in 2007 from Massey University, New Zealand (MUHEC Application 07/62). Pregnant women who were 16-years-or older were eligible to participate in the study. The ELF cohort recruited pregnant women at “parent and child shows” located in the main urban centers. Parent-child shows are large-scale events, marketed at expecting and experienced parents. People pay a small fee to enter these shows because it is a “1 stop shop” destination to purchase standard and newly available products (eg, food), services (eg, child-care), recreation and education programs (eg, developmental courses), and specialist advice (eg, child psychologist). The shows are attended by more than 22,000 people annually. The ELF study used other recruitment avenues including: information inserted in antenatal care booklets, promotional posters in hospitals and sonography clinics, and participants who enrolled through an Internet search engine. Thus, the study population included any expecting mothers, new and experienced, recruited through parent-child shows and other avenues, with access to the Internet. Participants were offered a “postal” option (offline) if they did not have access to the Internet, or if they preferred the offline option.

A final total of 2197 women were recruited in the study from September 2008 to September 2012. A large proportion of the participants were from Wellington (43.5%) with the other participants from Auckland (37.5%) and Canterbury (11.8%). A small proportion was from other regions (7.2%); and for less than 1% we had no current address. From the 2197 pregnant women recruited, 1,155 (52%) were categorized as lost to follow-up. The reasons were: (1) attrition to follow-up (81%); (2) participants who later declined to take part (12%); (3) missing information (1%); and other reasons (4%) such as miscarriage, nonviable pregnancy or death of the baby, and moved to another country and subsequently withdrew from the study. The final study sample analyzed in this paper is 1,042 participants. The majority of respondents (55%) took part via an offline mode, compared to 44.9% of online participants, and most women were recruited from the parent-child shows (73.2%), as described earlier.

The NINFEA and ELF studies are Internet-based cohorts examining protective and risk factors for a range of noncommunicable diseases in young children. These online birth cohort studies are the first of their kind in Italy and NZ. Both cohorts’ participants differed in age structure, with the largest group in the 32- to 35-year-old age group (35.7%) for NINFEA. The ELF cohort’s maternal age was predominantly younger (26- to 31-year-old age group). These age groups typically represent the median maternal age for both countries (NZ: 29 years; Italy: 31.3 years) [21,22], which explains much of the changes in the reproductive process (ie, birth delay) and stabilized fertility rates since the 1960s and 1970s [22. A comparison between the NINFEA cohort and the general population revealed that NINFEA participants are mothers with lower parity, higher education level, and lower frequency of smoking during pregnancy [16]. For the ELF cohort, online participants were notable by ethnicity and stage of pregnancy (trimester 3 having the highest participation for both online and offline). The latter characteristic falls in line with the NINFEA’s previous analyses, where women in their first trimester of pregnancy have a lower proportion of completed items from the baseline questionnaire [16]. Indeed, these findings are characteristic of Internet-based cohorts being a more self-selective sample of their respective source populations, and the timing (or in this case, the stage of pregnancy) for inviting participants to be take part in an online survey is an important consideration in order to attain complete responses and better respondent rates. Key risk factors of both the NINFEA and ELF cohorts indicated a reasonable comparability of participants who were smokers (8.4% and 8.6%, respectively) and drank alcohol (33% and 29%, respectively) during pregnancy, with clear differences between online and offline users in the ELF cohort. As more health outcomes data becomes available for the ELF cohort (ie, completion of subsequent phases), further analyses—including early life growth trajectory pathways to several health outcomes—will provide meaningful and useful interpretation.

The preliminary findings presented here show that an Internet-based cohort is feasible. Our investigation also highlights 3 major strengths that support the notion that Internet-based cohort studies are feasible and may have advantages in comparison to traditional cohorts: (1) given the prospective longitudinal nature of a cohort study design, an Internet-based approach can provide a significant research resource, particularly in the potential for expanding the breadth of a sampling frame and automated data downloading and cleaning that reduces the costs for administering a project long term; (2) the mode of Internet participation (eg, online questionnaires) has the potential to include multiple geographical sites for a long duration of time and to include large numbers of participants in the study (since recruitment through the Internet is less intensive as compared with traditional methods, an advantage of the Internet approach is that cohorts could recruit for many years; furthermore, there are provisions for identical cohorts to be established in other countries, and the online nature of the study could preclude additional costs for participation of mothers at an international level); and (3) as the protocols and online questionnaires are comparable for some phases of the ELF and NINFEA cohorts, particularly on future analyses on specific exposures and outcomes, this will allow for subsequent pooled analyses (these are currently being planned as the follow-up for each phase becomes more complete and for later phases when the children are of school age).

Limited access to the Internet, particularly for participants from a lower socio-economic background, may result in a selective cohort. Although this selection is not likely to introduce problems of validity in the associations measured, there may be issues of limited exposure heterogeneity in the study subjects. This would happen when the exposure of interest is strongly associated with participation and that there is limited variability in the exposure to investigate its effect on the outcome of interest. However, this problem is likely to be limited as in many countries, including Italy and NZ, the majority of the population not only has access to the Internet, but also access the Internet from a handheld mobile device such as a smartphone or an iPad (Italy: 58%; NZ: 88% in the whole population in 2012 and 2013 [23,24]). In addition, for some exposures, baseline selection may actually increase heterogeneities. For example, if high maternal age were the exposure of interest, having 25% of the cohort aged at least 36 years at delivery would increase the statistical efficiency of the cohort. Moreover, it is important to note that due to the Internet-based design and source population of the NINFEA and ELF cohorts, restricting the source population (like in our cohorts) are more likely to reduce issues of internal validity. This issue has been recently discussed the general consensus is that “representativeness” will depend on the context of a particular study, and thus it is a secondary issue [25]. Other researchers suggest that representativeness should be avoided, particularly if the study design incorporates an intentional nonrepresentative sample for practical reasons (eg, restricting the study to specific participants); minimizing bias by comparing subgroups; and if the focus was on 1 or more population subgroups [26]. This is the case for the cohorts currently presented in this paper, thus restricting the source population and internal analyses should not introduce serious issues of bias.

However, an important characteristic in all birth cohort studies where participants are followed-up is “attrition,” and we acknowledge that this is a particular issue for the ELF cohort. Anecdotally, ELF participants provided comments on addressing attrition or lost to follow-up. This included ideas of shorter questionnaires, reducing the interval time for the data collection phases, setting up electronic diary reminders with the participants, being very clear with participants to utilize the offline option if they are “not Internet-savvy,” and simplifying the Internet processes for enrollment (ie, there were some technical glitches that prevented participants from registering in a timely manner). These are important learnings from participants’ perspectives, and the authors accept that the points highlighted here should be considered for any future Internet-based research. Moreover, there is emerging work examining the follow-up of Internet-based epidemiological studies, and the findings advocate using an offline enrollment campaign as a potentially useful aid to achieve higher participation and to limit lost to follow-up. Based on the NINFEA and ELF cohort experiences, we cannot conclude whether attrition is higher or lower in Internet-based cohorts than in traditional cohort studies. As further phases are completed and the cohorts experiment with different online mechanisms (eg, use of social media tools), this issue will become clearer and will produce potential strategies to alleviate attrition at follow-up when using the Internet as a primary method of recruitment and data collection.

The initial work focused on the use of the Internet to conduct cohort studies.

First, some studies demonstrated empirically that baseline selection (or restriction) in cohort studies does not result in biased associations [16]. This has previously been recognized [27], and further support in the context of Internet-based research is needed. Data from the NINFEA study and the population-based birth registry of the Piedmont Region, Italy, were used to show that the confounding pattern in the NINFEA cohort differs from that of the general population, but this difference is not necessarily associated with a stronger overall confounding effect [16]. Simulation studies in which both the exposure of interest and an unmeasured strong risk factor for the outcome of interest, assumed to be independent in the general population, are strong determinants of the probability of participating in the Internet-based cohort were also performed, showing that even in the worst-case scenario, the magnitude of the bias introduced was small [16].

Further work to evaluate methods of recruitment for an Internet-based cohort and their potential effects on the study validity is ongoing. For example, the efficiency of a pilot advertisement campaign in Facebook, estimating a cost of €20 per participant, has been recently studied [30]. In addition, we found that both in the NINFEA and in the ELF cohort, the source of information (offline vs online) was associated with attrition at follow-up.

Studies on specific outcomes are also ongoing, in particular on growth in the first years of life and on wheezing. Data from the NINFEA cohort and 2 other non-Internet-based cohorts have been used to compare different approaches to model growth in the first 4 years of life [28], the association between several maternal prenatal exposures and weight trajectories in infancy were examined [17,28], and the paper highlighted a range of modeling options to estimate salient features of growth in weight in infancy and early childhood. However, the most useful was the SITAR (super-imposition by translation and rotation) model because of its flexible and pragmatic approach for life-course epidemiology inquiries. Finally, the NINFEA cohort participates in several collaborative studies among European cohorts, including those conducted under the CHICOS coordination project [29-34].

The 2 Internet-based cohorts presented in this paper had similar participant characteristics despite the differences in methods, data collection time frames, and source populations. Internet-based recruitment for epidemiological studies has the potential to expand a broader geographical coverage. However, online recruitment could introduce difficulties, particularly in the collection of biological samples, and it limits the capability to take standardized measurements (eg, weight, height). The NINFEA cohort protocol includes collection of saliva samples when children turn 6 months old, but it does not include cord or maternal blood sampling. There is, however, the potential of nesting ad-hoc studies in a subsample of the cohort to mitigate this issue.

There is much to learn about how to include the Internet as a valuable tool in epidemiological research. Over time, technological advances can only further aid in overcoming much of the current shortcomings, particularly in increasing follow-up and reducing the attrition rate. We encourage future studies to incorporate the Internet more strategically to decrease the limitations of individual and population-based approaches in epidemiological study designs.