Youth Experiences and Health Study

The current research will identify specific self-reported risk symptoms for psychosis and psychiatric hospitalization. Data will be collected in a large (N=1000) intake sample from the adolescent outpatients units of three Finnish university hospitals. Besides the central psychotic-like experiences, psychosocial protective factors of positive mental health, resilience, and social networks will also be simultaneously assessed, as well as potentiating risk factors such as trauma, sleep disorders, affective symptoms, borderline personality traits, and substance use. The predictive value of psychotic-like experiences, in isolation and in interaction with the other measures, will assessed with an inclusive national register follow-up of psychosis diagnoses and all-causes psychiatric hospitalization. The study will thus produce risk algorithms for everyday use in psychiatric care. For immediate applicability of results, tools for computerized self-report and automatic risk calculation will be made available for clinical use.

ANALYSIS PLAN To ensure a strong foundation for the primary predictor of positive psychotic-like experience (PLE) intensity, the structural validity of the YEAH questionnaire will be assessed with a theory-driven confirmatory bifactor model, taking into account the ordinal nature of items. In the case of unsatisfactory fit, exploratory analyses will be performed.

The main pre-planned outcome analysis will be baseline latent positive and negative PLE levels on the YEAH questionaire and their interaction predicting the follow-up transition to psychosis. Secondary analyses using the same predictors will use all-causes hospitalization as the outcome, as an indicator of a general deterioration of mental health. Exploratory LASSO analyses of the same outcomes will include as additional predictors all baseline assessments of psychosocial health and symptoms, as well as the symptom stability in the questionnaire follow-up. Best analysis and reporting practices recommended for the field will be followed (Studerus et al, 2017), to minimize chance findings and to allow accurate replication.

Power analyses The expected minimum sample size of 900 questionnaires is conservative for the planned main analyses. In our previous Helsinki Prodromal Study with a similar sampling strategy (Therman et al., 2014), 3.5% of respondents were diagnosed with psychosis within one year. With the consequently expected group sizes of 31 (psychosis) and 869 (others) and error probabilities α and β conservatively both set at 5%, a sensitivity analysis for a one-tailed t-test indicates a required effect size of 0.60 (Cohen's d) for the outcome-group difference in PLE intensity, which is the worst-case scenario. In our previous study, the d for the total score on the fairly non-specific Prodromal Questionnaire was 0.50, and the effect sizes for the brief questionnaires in the Kline et al. (2015) and Kobayashi et al. (2008) studies all corresponded to a d of 1.1 or greater. A d of 0.90 is a thus a conservative expectation with the new questionnaire, which would require a sample size of only 420. With a cut-off set to 20% test positives, already this effect size would achieve clinical usefulness as an initial screen, when the two-year psychosis rate is 5% (sensitivity 50%, positive predictive value 13%, diagnostic odds ratio 4.4). With a realistically obtainable d of 1.5 as the best-case scenario, the questionnaire screening would under the same conditions be highly useful (sensitivity 71%, positive predictive value 18%, diagnostic odds ratio 12).

DOCUMENTATION AND QUALITY Possibilities for continued use of the data will be ensured by carefully documenting data collection methods, external circumstances, as well as the contents of the dataset. For documentation, the project will use the Data Documentation Initiative (DDI, version 3.2 with lifecycle support) international metadata standard.

The primary database will contain a change log and previous versions will be archived with each revision. Main databases of the full dataset with version numbers and change logs will be maintained at the National Institute for Health and Welfare.

DATA CORRECTNESS The primary questionnaire data will be entered directly by the respondents as multiple-choice items on tablet computers. Aggregated and individual responses will be reviewed by clinical workers immediately afterwards and discussed with the respondents as part of clinical routine. There will therefore not be a separate data entry stage with associated data entry errors.

The secondary coded interview data will be entered by clinical workers directly into the database, and will be reviewed by the treating physician as part of clinical practice. Entry fields will be multiple-choice or limited to allowed values.

Correct identification of data will be ensured by the clinical worker entering as primary identifier the unique Finnish national identification number, with the full name as secondary information. The identification number is subject to immediate automatic checks using the builtin check character and the associated age, and it will later be cross-checked with the name in local register follow-ups.

Imported register data will be screened for disallowed values, and dates cross-checked with the primary study database to ensure correct person identification. Local and nationally aggregated register data coverage will be compared to ensure completeness.

ADDITIONAL SECURITY OF PERSONAL IDENTIFYING INFORMATION Personal identifying information will be stored separately from other electronic data, and linked to clinical data by a new random study identifier (a uuid) generated at THL. A uuid, or universally unique identifier, is a 128 bit number, ISO/IEC 9834-8:2014). All analyses will be conducted using datasets without other identifying information than the study uuid. The file with personally identifying information will additionally be encrypted with a study-specific password known only to participating researchers.

DATA SHARING AND LONG-TERM PRESERVATION All materials will be permanently stored in the THL archives as per the Finnish Archive Law (23.9.1994/831).

To ensure maximum discoverability, metadata including codebooks and employed analysis scripts will be archived at the Finnish Social Science Data Archive (FSD) and within the EUDAT metadata infrastructure B2FIND.

Open sharing of the data is not possible as consent for publication of the raw data is not obtained from the participants and the dataset could pose a threat to the confidentiality of their sensitive information.

Due to the sensitive nature of the data and commitments made to the participants on their consent form, the compiled data itself will be managed solely by the National Institute for Health and Welfare (THL) and will be made available to other parties only by separate agreement and in anonymized, redacted form. This redaction work will be in compliance with national guidelines and additionally follow the most restrictive recommended approach among the following authoritative resources:

1. US National Institutes of Health Data Sharing Policy
2. UK Data Archive Anonymization/Quantitative
3. Standards published in leading medical journals (e.g. BMJ 2010;340:c181, doi: 10.1136/bmj.c181)