Sample

The FTC was established in 1974 and is composed of twins born in Finland before 1958 (full range of birth years: 1880−1957). There have been four waves of questionnaires, conducted in the years 1975, 1981, 1990, and 2011. The FTC originally included 13,888 same-sex twin pairs of known zygosity in 1974 and expanded in 1996 to include approximately 8000 pairs of opposite-sex twins born between 1938−1958. Individuals were invited to participate in later waves irrespective of participation in earlier waves. Zygosity of same-sex pairs was determined by validated questionnaire in the 1975 survey, with validation via genetic markers conducted for those twins who have also participated in studies of selected twin pairs. There remain 889 same-sex twin pairs in the FTC for whom zygosity is undetermined due to inconsistent questionnaire responses. The number of pairs includes both complete and incomplete twin pairs; in the 1975 survey 83% of all pairs are complete twin pairs.

In addition to the FTC, we obtained register data from the spouses of the FTC, siblings of the FTC, and spouses of the siblings of the FTC. The basis for the TwinRegistry data began from the FTC and the sample was expanded to facilitate various specific research designs. We additionally obtained intergenerational data by including the parents of the FTC, the children of the FTC, and the children of siblings of the FTC. The FTC questionnaires are linked with the register data via personal identity code assigned to Finnish citizens and residents.

Lastly, we created a reference population composed of all individuals born in Finland between 1945−1957. This reference population aligns with a subset of the FTC that were eligible for inclusion in all four FTC questionnaires (birth years 1945−1957 eligible for surveys in 1975, 1980, 1990, 2011). Then, for the reference population, we obtained their family members for the same relationships included for the FTC (e.g., spouses, parents, children, siblings, siblings’ children). The familial linkages include both biological and adoptive family members (parents, siblings, and children); spousal relationships are recorded on a yearly basis.

Across the FTC, reference population, and all linked relatives, TwinRegistry includes approximately 4.6 million unique individuals. A single individual can have multiple familial roles. Figure 1 provides an illustration of the sample and familial relationships. From these linked familial relationships, we can create complete familial units; for example, complete parent-child trios for the FTC and reference population. Intergenerationally, there are approximately 25,000 complete parent-offspring trios from the FTC and their parents, 840,000 from the reference population and their parents, 54,000 from the FTC and their children, and 980,000 from the reference population and their children. Table 1 and Table 2 provide a more thorough overview of sample sizes for various roles (Table 1) and familial units (Table 2).

Figure 1.

Figure depicting multigenerational data available from the Finnish Twin Cohort and register sources.

Table 1.

Sample sizes by role

Note: An individual may have multiple roles. For example, a sibling of a twin could themselves be part of the reference population; the biological child of a twin mother may also be a biological child of a reference individual father.

Table 2.

Sample sizes for various family units

Note: Biological and adoptive families are combined.

The following sections on register and questionnaire data provide a detailed narrative on a collection of comprehensive datasets available in TwinRegistry, describing aspects of social, economic, and health data. These datasets serve as resources for researchers exploring longitudinal trends in health, demographic changes, and the impact of policy interventions over time. All data use pseudonymized IDs for confidentiality.

Register Data

TwinRegistry data from the registers can be categorized into several partially overlapping categories: demographics, education, employment and socioeconomic position, health and social care, benefits and pensions, and criminal offences (see Table 3 for details). Registers have varying years of data collection and come from several different Finnish agencies. Ready-made datasets, compiled by Statistics Finland, are available for the entire Finnish population, as stated in the metadata as provided by Statistics Finland. Custom datasets, or those created specifically for the TwinRegistry project, are available for the subsample of FTC individuals, the reference population, and all linked relatives (∼4.6 million individuals). Sample sizes for the registers vary based on the year of data collection and whether the dataset is ready-made or custom. The sample sizes for the number of individuals range from 91,909 (custom-made KELA pension dataset) up to 8,552,770 (ready-made FOLK children and parents’ data). Similarly, some registers may contain multiple observations for a single individual, and these sample sizes also vary substantially by register type and number of years of data collection. The smallest numbers of observations range from 1,048,575 for the custom-made KELA pension data and up to 211,555,041 observations for the KELA dispensed prescription register. Any diagnostic data is coded via International Classification of Diseases codes (ICD; ICD-10 Version: 2019, n.d.), though the version of ICD relevant depends on the register and date, as described in available metadata from the agencies hosting each register. Prescription data for medications uses Anatomical Therapeutic Chemical (ATC) classification codes.

Table 3.

Domains of register data, specific registers included, years of coverage, and example data

Questionnaire Data

The FTC questionnaires offer longitudinal insights through surveys conducted in 1975, 1981, 1990, and 2011. In 1974, researchers identified potential twins from the Central Population Registry of Finland, selecting pairs of individuals of the same sex with the same birth date, same surname at birth, and same local community of birth (Kaprio et al., Reference Kaprio, Bollepalli, Buchwald, Iso-Markku, Korhonen, Kovanen, Kujala, Laakkonen, Latvala, Leskinen, Lindgren, Ollikainen, Piirtola, Rantanen, Rinne, Rose, Sillanpää, Silventoinen, Sipilä and Waller2019). The initial survey in 1975 was sent to these identified individuals and inquired whether the individuals were twins, as well as zygosity information for twins. A total of 36,922 individuals were invited in 1975, of which 31,145 replied (response rate of 84.6%). This number of responses includes N = 1175 individuals who indicated they were not a biological twin. A total of 29,311 individuals were invited in 1981 with a total of 24,684 responses (84.2% response rate), including 2451 individuals who had not responded in 1975. In terms of retention, 22,122 twins participated in both 1975 and 1981, with 4308 twins who participated in 1975 but not 1981 (83.7% retention from 1975 to 1981).

In 1991 and 2011, only individuals in certain birth years were invited to participate. In 1991, 16,179 twins born between 1930−1957 who had also participated in either 1975 or 1981 were invited and the response rate was 77.3% (N = 12,502). In 2011, all twins born between 1945−1957, regardless of earlier participation, were invited to participate (N = 11,738 invited, N = 8410 responses, participation rate 71.6%).

These surveys cover aspects of health, employment, family dynamics, personality, stressful life events, and housing conditions, reflecting changes in respondents’ situations over the life course. As the FTC questionnaires have been described extensively elsewhere (Kaprio, Reference Kaprio2006, Reference Kaprio2013; Kaprio et al., Reference Kaprio, Bollepalli, Buchwald, Iso-Markku, Korhonen, Kovanen, Kujala, Laakkonen, Latvala, Leskinen, Lindgren, Ollikainen, Piirtola, Rantanen, Rinne, Rose, Sillanpää, Silventoinen, Sipilä and Waller2019; Kaprio et al., Reference Kaprio, Sarna, Koskenvuo and Rantasalo1978; Kaprio & Koskenvuo, Reference Kaprio and Koskenvuo2002), here we describe the broad data domains (see Table 4 and the supplementary table) and highlight the evolution of the surveys in content and methodology, reflecting shifts in research focus and public health priorities.

Table 4.

Domains of Finnish Twin Cohort questionnaire data

Broadly speaking, the domains included in the questionnaires remained stable over time, but the degree of detail varied with topical research interests and the sample age. For example, items on weight and body measurements became more detailed over time, reflecting a growing concern with longitudinal weight trends and waist circumference as a measure of abdominal obesity. Types of diseases, medications, and chronic health conditions included over time also became more inclusive and detailed, reflecting new knowledge about disease as well as the aging cohort. Health-related impacts of sleep, substance use, pain, and psychological health also became more expansive over time. While many questionnaire domains became more detailed over time, many items were maintained over time (e.g., the alcohol and tobacco use items; see the supplementary table), allowing for consistency in longitudinal analyses.

Data Resource Use

The linkage of FTC questionnaires with register data expands opportunities that utilize classical twin designs and discordant twin designs, by providing both novel exposures and outcomes, and a follow-up time of several decades. The additional availability of the reference population from the same birth cohort as a subset of the FTC permits a thorough study of generalizability from twins to the broader population, as well as the use of other genetically informative family-based research designs such as sibling and cousin analyses in practically whole-population data. The combination of twin designs with individual level analyses will permit for excellent statistical power and controls for confounding factors. Additionally, multigenerational studies are also feasible here, including the children of twins design (McAdams et al., Reference McAdams, Hannigan, Eilertsen, Gjerde, Ystrom and Rijsdijk2018) and other extended twin family designs. These extended familial designs permit analyses of intergenerational transmission, better estimate nonadditive genetic effects, and estimate the effects of assortative mating and parental environmental effects (Keller et al., Reference Keller, Medland and Duncan2010).

TwinRegistry is currently developed by researchers at the Institute of Molecular Medicine Finland (FIMM) and the Institute of Criminology and Legal Policy (Krimo) at the University of Helsinki. FIMM notably also hosts the FinnGen and FinRegistry projects described elsewhere (Kurki et al., Reference Kurki, Karjalainen, Palta, Sipilä, Kristiansson, Donner, Reeve, Laivuori, Aavikko, Kaunisto, Loukola, Lahtela, Mattsson, Laiho, Della Briotta Parolo, Lehisto, Kanai, Mars, Rämö and Palotie2023; Viippola et al., Reference Viippola, Kuitunen, Rodosthenous, Vabalas, Hartonen, Vartiainen, Demmler, Vuorinen, Liu, Havulinna, Llorens, Detrois, Wang, Ferro, Karvanen, German, Jukarainen, Gracia-Tabuenca, Hiekkalinna and Ganna2023). These three similar initiatives each have unique samples, strengths, and capabilities, but they do all share register data as a basis, which allows for coordination and infrastructure sharing across projects. For example, the FinRegistry project has compiled Risteys, a publicly available web portal that enables exploration of the over 3100 clinical endpoints available in FinRegistry (Risteys, n.d.). As the underlying register data in TwinRegistry and FinRegistry are from many of the same sources, many of the FinRegistry endpoints can also be generated in TwinRegistry.

As an example of causal analysis based on the register data, our recent work evaluates the influence of job loss on all-cause and specific-cause mortality, utilizing the economic job displacement natural experiment at the individual-level (Zellers et al., Reference Zellers, Azzi, Latvala, Kaprio and Maczulskij2025). Finland suffered a severe recession in the early 1990s, with numerous mass layoffs and plant closures resulting in high unemployment. These job losses due to plant closure and mass layoff can be identified objectively via register data, highlighting one example of the utility that this administrative data provides. Future work will integrate the behavioral data available in the twin questionnaires to evaluate lifestyle factors as possible mechanisms from job loss to mortality, as well as leverage ‘children-of-twins’ designs to investigate how parental job loss influences offspring mental health.

Generalizability

The comprehensive linkage of FTC data to registers and availability of register data in the reference population born 1945−1957 permits for thorough analyses of how appropriately twins participating in studies represent their birth cohort. Future projects will directly evaluate representativeness as is relevant to the projects’ specific research questions, but here we present a starting set of analyses to demonstrate the representativeness of twins to their birth cohort. To assess representativeness, we compared the subset of FTC twins born 1945−1957 that participated in the 1975 survey (N = 23,650) to the singletons in the reference population born the same years (N = 1,254,094) using register variables anticipated to be useful in a variety of projects. As the full reference population does include those FTC members born 1945−1957, all comparisons between the FTC sample and singleton members of the reference population are conducted after removal of the FTC from the reference population.

Here we visualize representativeness of the FTC on education, mortality, employment, residence, and marital status. We draw data from the 1987 FOLK basic data (first year of availability) and the 2022 FOLK basic data (last year of availability). Results are presented in Table 5 (1987) and Table 6 (2022). We present relative proportions rather than statistical comparisons, as the sample sizes are well powered to detect small but statistically significant effects that may not be meaningful to interpret. Generally speaking, the twin sample is not different from their respective birth cohort, but this will be more thoroughly evaluated in later papers, with respect to the key outcomes and ages of interest, including statistical comparisons to quantify generalizability and statistical significance. See Niemi de Paiva et al. (Reference Niemi de Paiva, Hukkanen, Kaprio and Zellers2025) for an example of ongoing generalizability analyses. We note there are N = 406,891singletons (32.4% of the full reference population) who are without any siblings. This is a special group that necessarily can only occur in the singleton population and may influence differences, though we do not evaluate this here.

Table 5.

Generalizable descriptives based on 1987 FOLK basic data, comparing twins born 1945−1957 and population reference individuals born same years (age range 30−42)

Note: Individuals may be excluded from the FOLK basic dataset for a given year if they were not living in Finland on the last day of that year due to emigration or death (e.g., 31 December 1987).

Table 6.

Generalizability descriptives based on 2022 FOLK basic data, comparing twins born 1945−1957 and population reference individuals born same years (age range 65−77)

Note: Individuals may be excluded from the FOLK basic dataset for a given year if they were not living in Finland on the last day of that year due to emigration or death (e.g., 31 December 2022).

Strengths and Limitations

A major strength of the TwinRegistry is the ability to combine register data with self-report data, as each data source has its strengths and limitations. The register data is well powered and contains a wealth of objectively reported clinical phenotypes but lacks information on behaviors and lifestyle that may be relevant for less severe outcomes. For example, register data would contain information on diagnoses of alcohol use disorders, which are given to patients whose symptoms are severe enough to warrant medical diagnosis and intervention, but individuals with subclinical alcohol use problems would not be well characterized in register data. On the other hand, in the FTC questionnaires, we have information on alcohol consumption, which is useful for investigating normative use, but since the FTC is a community sample, severe and heavy use are rarer and therefore challenging to investigate using only questionnaire data. The combination of register and questionnaire data provides complimentary information across the spectrum of severity. Questionnaire data additionally provides an advantage in that they contain information on risk factors, exposures, and symptoms not available in register data.

Also, longitudinal studies requiring active participation are limited by potentially losing some of the most important subpopulations during follow-up, such as people who develop alcohol use problems, whereas register data has no such issues with attrition. Using both data sources simultaneously permits a nuanced set of analyses on normative, subclinical, and clinically relevant outcomes. An additional major strength is that register data is free from recall bias, has no loss to follow-up, and is sufficiently detailed for long-term studies.

A limitation with respect to data utility concerns biological data. As part of the Finnish Biobanking Legislation (Law 688/2012), all biological samples associated with the FTC were deposited in 2018 with the nationwide Biobank of the National Institute for Health and Welfare (Kaprio et al., Reference Kaprio, Bollepalli, Buchwald, Iso-Markku, Korhonen, Kovanen, Kujala, Laakkonen, Latvala, Leskinen, Lindgren, Ollikainen, Piirtola, Rantanen, Rinne, Rose, Sillanpää, Silventoinen, Sipilä and Waller2019). Biobank data can be linked to FTC questionnaires or to register data, but so far, it has not been possible to combine biological samples with FTC questionnaires and register data simultaneously. Therefore, projects can link two of the three data sources (register data, biobank data, questionnaires), but not all three, to protect participant privacy, and thus the TwinRegistry does not currently contain biological data. An additional limitation is that the full reference population is only available for birth years 1945−1957, therefore we are only able to conduct full-population generalizability analyses for the birth years 1940−1957. Generalizability of the twins born 1880−1944 to their full birth cohort cannot be comprehensively assessed. That said, we have no reason to believe that generalizability of twins to singletons varies by birth year on the outcomes described here.