Study protocol for Controversies in psychiatry – antipsychotics: Long term outcomes and causal modelling of antipsychotic treatment using Norwegian Registry data

Controversy

The situation can be depicted as a controversy where practice and recommendations can be arranged on a dimension from liberal to restrictive. The liberal position is concerned about under-treatment, and those concerned may recommend early initiation, higher or stable doses, or longer or permanent maintenance treatment, and warn strongly against practices such as treatment without antipsychotics. In contrast, the restrictive position, concerned about over-treatment and side effects, advocates for lower doses, shorter maintenance periods, or more frequent acceptance of care without antipsychotic treatment.

Controversies tend to fade when faced with convincing evidence, particularly from randomized controlled trials (RCT), which excel in acceptability and feasibility for short to intermediate causal effects of medications. In antipsychotic care, the controversy partly concerns long-term maintenance treatment and the feasibility and safety of managing psychotic disorders without starting antipsychotics at all. Randomizing to care without medication has been challenging due to ethical and practical considerations.^37,38

Randomization for extended exposure time raises ethical concerns, increases trial costs, and complicates recruitment and retention. Strong patient opinions and difficulties in recruiting participants willing to consent to randomization have posed challenges in recent studies on the gradual discontinuation of antipsychotic treatment.³⁹ Some restrictive proponents have also argued that RCT evidence where patients are blinded to stable dose or a more or less abrupt discontinuation without compensatory measures, have limited relevance for patients that carefully plan to gradually discontinue antipsychotics.⁴⁰ Furthermore, potential selection bias is a significant issue in RCT samples of individuals with psychosis, as 80% of the general care population is excluded due to elaborate inclusion and exclusion criteria.⁴¹ Making use of nationwide registry studies for causal analysis could mitigate these issues and provide pertinent evidence.

When care ingredients vary between providers or areas, causal analyses with registry data may be possible. A recent mental health atlas reported large variations for several care parameters in Norway.⁴² Consequently, we anticipate variation in relevant parameters for this study, such as prescribers’ preference for type, dosage and length of antipsychotic treatment. We also expect that some providers give patients strong warnings against treatment outside guidelines, such as not using antipsychotics, while others will be more tolerant to such patient preferences.

Aims

This paper presents an overview of how we aim to study the relation between use of antipsychotic medication and future social outcomes and health care use, following an inpatient episode or a new psychotic episode diagnosed in outpatient care. Specifically, we aim to address the following research questions:

These questions aim to provide comprehensive insights into the effects of different treatment approaches on the well-being and outcomes of individuals diagnosed with psychosis.

Setting and context

The study is set in Norway, a sparsely populated northern European country with approximately 15 inhabitants per square kilometer. The public health care system is predominantly funded by tax revenues and serves as the primary provider of services for severe mental disorders, with few alternatives existing outside this single-payer system. Norway benefits from several comprehensive health, care, demographic, and social registries, which utilize unique social security numbers to facilitate highly complete and linkable data.

Long distances and mountainous terrain have stimulated regional and local adaptation and self-determination, and recruitment of qualified staff is difficult in some areas. We therefore expect local providers to vary in their care preferences, and that this also influences antipsychotic treatment decisions. Which provider you happen to get, can therefore set the stage for a lottery-like situation for parts of the treatment choice, in addition to variation predicted by patient characteristics such as symptom load and treatment history.

Study population and data acquisition

The study will have access to data for all patients aged 10 years and older who received specialist mental or addiction health care between 2015 and 2016, estimated to be approximately 300,000 individuals. These patients will be identified through the Norwegian Patient Registry (NPR), which contains reliable, patient-identifiable national data for specialist inpatient and outpatient care with high completeness.⁴³ A similarly sized control cohort matched on age and sex will be drawn from the general population by the National Population Register. For both the patient and control cohorts, we will borrow individually linked data on their health care use and social status from the age of 10 onwards, for the years 2010 to 2025, from the Norwegian Patient Registry (NPR), Norwegian Prescribed Drug Registry (LMR), Central Population Register (CPR), Norwegian Registry for Primary Health Care (KPR), Norwegian Cause of Death Registry (DÅR), Norwegian Control and Health Reimbursements Register (KUHR), Central Penal and Police Register (SSP), Norwegian Education Database (NUDB), and Tax, Income and Welfare Registry (FD-trygd). The structure of data in time is illustrated in Figure 1. Diagnoses for severe mental disorders in the NPR are found to be valid,⁴⁴ and consultations, admissions and collected prescriptions, which are reimbursed, are registered with high accuracy. This protocol regards the subsample of patients diagnosed with non-affective psychotic disorders (ICD F20-29). In 2015, the number of patients with these disorders aged 18-65 in Norway treated in specialist mental health care was 11,000.⁴⁵ Analyses for diagnostic subgroups, such as schizophrenia (F20), can be conducted when the sample size permits. Data for adolescents under 18 and for at least five years of pre-exposure (2010-2014) will be available, allowing us to distinguish incident from prevalent cases and to control for some pre-exposure function and treatment variables.

Figure 1. Overview of available variables from Norwegian registries for Controversy in Psychiatry – Antipsychotic medications.

Exposures and outcomes

We will define exposures using data from the Norwegian Prescribed Drug Registry. Table 1 provides an overview of the planned exposures and outcomes. Antipsychotic use will be classified based on the Anatomical Therapeutic Chemical (ATC) classification N05A⁴⁶ excluding lithium. We will estimate the consumption of collected prescriptions using the Prescription Drug Purchases to Drug Use Periods (PRE2DUP) method.⁴⁷ Initial outpatient antipsychotic use and non-use will be determined after a 30-day period following discharge, as in J Tiihonen, A Tanskanen and H Taipale¹¹ or a similar period following an outpatient consultation where the patient has been given a new non-affective psychotic disorder diagnosis.

Nordic prescription registries lack data on hospital medications.⁴⁸ This limitation hinders the ability to detail medications administered in hospital or clinic settings, for example during admissions or under community treatment orders. It is reasonable to expect that the majority, but not all, patients in such care are treated with antipsychotics.^49,50 Consequently, if a patient does not collect any antipsychotic prescription after discharge or revocation of their community treatment order, it often signifies an abrupt discontinuation of antipsychotic medication. Therefore, this study will focus on exposures in outpatient care or make assumptions regarding inpatient antipsychotic use.

The dataset will include data on various outcomes such as injuries and deaths, emergency room visits, mental health diagnoses, inpatient and outpatient care, involuntary care status, educational attainment, occupational income, welfare reliance, supported housing, and crime. Some of the functional outcomes, such as education, work participation, welfare benefits and deaths, are mainly relevant as intermediate and long-term outcomes (see Figure 1).

For all exposures and outcomes, we will report results from observational analyses comparing patients with and without the specified exposure, as well as comparisons with the control cohort. We will control for pre-exposure indicators of severity and functioning, including prior specialist mental health care for non-psychotic disorders, injuries, education, and income. Additionally, variations in public purse and societal costs dependent on exposure can also be estimated, with appropriate local unit costs attached to resource use, and results reported in a common price year.

Instrumental variable analyses

We plan to estimate the causal effects of the antipsychotic exposures listed in Table 1 using variation in prescribers’ preferences as an instrumental variable. For the use of a particular antipsychotic substance, dose, or treatment duration, the instrument will be constructed based on each prescriber’s empirical frequency of administering these treatments. We will use the leave-one-out mean to ensure independence between the patient and the instrument.⁵¹

The extent of variation in Norwegian prescribers’ preferences for the exposures in Table 1 is currently unknown and must be examined empirically. Given the situation and controversies outlined in the introduction and methods sections, along with the lack of ubiquitous treatment guidelines for several exposures, we expect sizeable variation in these exposures among prescribers. This variation should result in a strong correlation between the instrument and the exposures, which is important to avoid bias⁵² and ensure a valid instrumental variable analysis.⁵³

We expect most patients to be unaware of professional disagreements, thus remaining blind to local or individual practice variations. Consequently, it is unlikely that patients can select prescribers based on accurate knowledge of individual prescription patterns. Prescribers’ use of various exposures will also be partly influenced by their case mix. Therefore, we will include baseline covariates such as demographics and pre-inclusion mental health treatment in our analyses.

Two other important assumptions for instrumental variable analysis that must be theoretically justified are exclusion restriction and non-confounding. Exclusion restriction requires that the instrument influences the outcome only through the exposure and not through other causal pathways. We expect that the propensity for the antipsychotic exposures does not affect outcomes through other causal paths.

Psychosis care without antipsychotics will be examined at the level of community mental health center catchment areas because our registry data will not allow us to allocate an individual prescriber to a patient who does not collect prescriptions. If exclusion restriction or non-confounding of the outcome is violated for care without antipsychotics at the community mental health center level, we anticipate that the instrument will predict a specific outcome beyond the patient sample. To explore this possibility, we will employ the zero-first-stage test on the control population.⁵⁴

For non-confounding, we will investigate observed variables that could potentially confound the relationships between the instrument and outcome. Several of these variables will also be controlled for as baseline covariates. If the conditions necessary for instrumental variable analyses are uncertain or not met, alternative analytical approaches will be considered, such as hierarchical models and inverse probability weighting.

The power of the instrumental variable analysis will depend on several currently unknown parameters, including the magnitude of relevant covariates and the variation among prescribers for each exposure. MA Hernán⁵⁵ argued that in causal analysis with observational data, the objective is to estimate a range of effect estimates that are highly compatible with data. Consequently, even studies considered “underpowered” that yield a broad range of estimates should be conducted and published. This approach makes results available for future meta-analyses aimed at reducing uncertainty by integrating data from multiple sources.

Public and patient involvement

The study is presented at https://www.helse-bergen.no/avdelinger/psykisk-helsevern/forskingsavdelinga-divisjon-psykisk-helsevern/populasjonshelse/antipsykotika/. The exposures and outcomes detailed in Table 1, along with the broader analysis plan, have been discussed with individuals with lived experience on several occasions during the project’s development and planning phases. The current paper has undergone critical review by a co-author with lived experience (AB).

Ethical approval

The study protocol has been approved by the Regional Committee for Medical Research Ethics South East Norway, Committee D (REK 10895), from 2022 to 2030. The committee determined the project to be of significant utility, ensuring the welfare and integrity of participants.

Patient consent and confidentiality

The study will use registry data collected for administrative purposes. According to Norwegian legislation, permission can be given to research institutions to borrow and analyse de-identified data from existing health registries and other registries without individual consent, as long as participants’ welfare and integrity are maintained.

Individual informed written or oral consent was deemed to be infeasible by the Regional Committee for Medical Research Ethics South East Norway. The Norwegian Directorate of e-health has approved the use of pseudonomized data for the aims of the project (H-398/S-84DD4B7E). A Data Protection Impact Assessment (DPIA) has been conducted and will be updated as necessary. The study adheres to the Declaration of Helsinki. When this manuscript was first submitted, we were still awaiting data delivery.

The NPR will identify the cohort, while Statistics Norway will draw a control cohort from the remaining population, matched by age and sex, and assign each person a project-specific pseudonym number. Various registries will lend out data based on social security numbers; however, the researchers will only receive pseudonymized data. Statistics Norway will retain the linkage key for a second wave of outcome data. The project will securely store and analyze data on protected servers, accessible only to researchers approved by the Norwegian Regional Ethics Committee.