Adaptation of Healthcare Services for Home-Dwelling Individuals with Dementia: A Programmatic Registered Report Using National Registry and Caregiver Data

Manuscript title: Adaptation of Healthcare Services for Home-Dwelling Individuals with Dementia: A Programmatic Registered Report Using National Registry and Caregiver Data
Manuscript link: https://osf.io/9s5zx/files/osfstorage
Invitation to Review date: 12.09.2026
Review submission date: 16.09.2026
Review type: original submission

I was invited to review the manuscript “Adaptation of Healthcare Services for Home-Dwelling Individuals with Dementia: A Programmatic Registered Report Using National Registry and Caregiver Data” with a focus on the overall study design. While this is not my area of expertise, I agreed but want to emphasize that I am no expert on longitudinal studies or dementia.

The authors describe a detailed plan for three studies on home-dwelling individuals with dementia and how their health is changing, how the services they use might be changing, and how informal caregivers perceive this change. I think that the study sounds relevant and interesting, though I have no expertise in the topic and cannot evaluate its relevance. I further liked the programmatic approach, the combination of different methods, use of register data, and the summary in Table 3.

Evaluation: For context, my understanding of preregistrations and registered reports is that they deprive researchers of all their degrees of freedom. They are the answer to the problem of undisclosed flexibility (10.1080/19312458.2015.1096329) and I have made the point that this flexibility goes far beyond optional stopping and the mere absence of preregistered code can introduce degrees of freedom that inflate the alpha level (10.31222/osf.io/v259t_v2). Against this backdrop, I recommend a specification of the entire analysis plan. A preregistered analysis script that runs on mock-data, a very small subset of the data, or simulated data would solve most of the issues that I am raising. Subsequently, a reviewer who is experienced with longitudinal multilevel models or time-lagged models should review the analysis plan. This lack of specificity may be appropriate for a normal research paper but I do not think that it works as a registered report as currently written.

Lukas Röseler

Remarks in no specific order

1. Spell out KOSTRA in the abstract like the other acronyms
2. “population-based longitudinal evidence“ sounds weird, I suppose the evidence is not longitudinal but the study design upon which the evidence is based
3. I found the first six pages or so very overwhelming, with lots of evidence on dementia in a semi-structured way. While I do not have a good suggestion myself, maybe there is a way to structure this part more and provide either a clearer theoretically oriented derivation of the hypothesis or a practical one.
4. Methods
   1. Maybe this is just because I am no expert on the topic but I was wondering about the eligibility criteria: Are you doing a mortality analysis / condition on survival through follow-up since you require individuals to “have available registry data … during follow-up”?
   2. You wrote “Dementia will be identified using registry-based diagnostic information from the Norwegian Register for Primary Health Care (NRPHC) and, where relevant, the Norwegian Patient Registry (NPR)”. What are the cases where NPR is relevant?
   3. Preregistered analysis script: I recommend to only accept a stage 1 Registered Report (or take a preregistration seriously) if the analysis script is also preregistered (e.g., with mock data or simulated data). Please provide an analysis script. I do not think that the description of the analyses rules out all degrees of freedom (see my argument here 10.31222/osf.io/v259t_v2). For example, the order of your exclusion criteria may be different in the script, which could affect the sample and some criteria could be unspecified, leaving degrees of freedom.
   4. Can you explain what the scope of the registers is? Is it containing the whole population? If so, you may discuss why you use inference statistic and what you want to generalize to (individuals in other countries, future individuals, …).
   5. power simulation / sensitivity analysis: I do understand that you did not conduct a power analysis. However, you should run sensitivity analyses and determine relevant effect sizes (e.g., plot a curve of the smallest findable effect size but also determine what effect size would be practically relevant). Also, what do you mean with “large cohort”? In EEG research, 50 individuals can be considered large, whereas in sociology, 50K individuals may be considered large (this would maybe be irrelevant if the journal that this would eventually go to was specified but with PCI-RR this is not the case).
   6. Computation of variables: You should specify what “indicators will be derived from the available ADL data“ means. Will you sum 0-1-values up, standardize them somehow, are they processed already, etc.? “according to registry documentation“ -> Please add a citation of the documentation – ideally this should be highly specific. Again, an analysis script would resolve many of these issues.
   7. Statistical model: “receipt of home nursing services and allocated service hours per week” This sounds like a whole number that likely has a skewed distribution. You later describe a longitudinal mixed effects model but do not say if or what link function you plan to use. Later, you also have outpatient specialist healthcare utilization as another number (“number of specialist outpatient contacts during follow-up”.
   8. “Where available, measures reflecting the scope of service provision (e.g., multiple service providers involved in care) will be considered in supplementary analyses.“ -> You should move this to exploratory analyses since it is very unspecific. Similarly, please clarify what role the secondary outcomes play (p. 13 and also Table 1). How are the results relevant for the interpretation of the primary indicators?
   9. “Random-intercept models will be employed, with random slopes for time considered where appropriate.“ (later “where data permit.”) I think that you should already know where this is appropriate or define formal conditions that determine based on the data where it is appropriate.
   10. data access before preregistration: Please explain whether you have worked with the data before and whether you have had access to the full or partial dataset.
   11. Computation of central variables: You explain (p. 10) that “together, these registers provide information…”. Does that mean that they provide complementary variables or are you computing variables based on the combined datasets?
   12. Specificity of hypotheses: H1c does not say how closely the declines will be followed by increases in home nursing utilization and does not provide a reference point (e.g., that they will not affect other types of service utilization)
   13. Effect size estimation: H2a says that there will be a significant relationship, though with large incidental dataset, the crud factor argument (10.2466/pr0.1990.66.1.195) applies, where everything is related significantly in large datasets. H2b does not specify what variation means (e.g., larger than 0 standard deviation, meta-analytical heterogeneity) and should have a realistic reference point (i.e., one that is not zero).
   14. Pre-specification of interpretation of hypothesis-(in)consistent outcomes: It is unclear what specific data patterns would support versus disconfirm the hypotheses, or leave the research question unanswered. I very much like your listing in Table 3 but there the interpretation if not supported is very vague and I think you could prepare further analyses to make such an outcome more informative. For example, in row 1 of Table 3, you say that an unsupported effect may indicate selection effects, whereby you admit that you would not see H1a disconfirmed. So when would you consider H1a disconfirmed? It would be a shame if such a complex study would not be able to falsify your beliefs.
   15. “Records identified as administrative duplicates or containing implausible values that are inconsistent with registry documentation will be excluded if detected during data processing.“ 🡪 This is another case of a post hoc analysis that you would essentially be able to decide however you like. Implausibility should be defined a priori. If you have to deviate from that definition because you did not expect a specific pattern that is fine and would require a discussion of deviations from the preregistration.
   16. Justification for the qualitative study to come after the quantitative studies: Usually, research is starting in an exploratory fashion, generating hypotheses that are subsequently tested (confirmatory research). I recommend you to explain the order of the studies more explicitly.
   17. Data availability: Please explain how the data can be accessed and, in the case that others seek to reproduce your analyses, what permissions they need to get from where.
   18. Missing data: I expected register data to be a case where you check if the data is/are missing at random and maybe do an imputation. Maybe consider this as an option and consider what proportion of missing data you expect.
   19. Tests of requirements for analyses are missing. You plan to compute ICCs for H2a but do not say what you plan to do if they are extremely small or large.
5. Study design: I think it would maybe make more sense to have a clear comparison group to make sure that both declines (that you argue follow each other) are not a general trend. This could be a matched cohort of home-dwelling older adults without registered dementia to check if any trends are specific to dementia.
6. I recommend either checking (if you have not done so) and referring to standards on registering qualitative study or more clearly labelling Study 3 as exploratory. For example, you could already describe the entire semi-structured interview guide.
7. I recommend not accepting the study unless it has received final approval from the ethics committee.
8. I note this is submitted as a Programmatic Registered Report, which permits separate Stage 2 outputs. I suggest that you state the expected sequence and timeline of the three Stage 2 submissions, a commitment to submit all three regardless of outcome, and how Stage 1 specificity will be preserved if Study 3 is conducted years after IPA.
9. If an ADL reassessment is triggered by an application for (or revision of) services, then observing that "ADL decline is followed by an increase in home nursing hours" partly documents the administrative workflow rather than the responsiveness of the care system.

Disclaimers

1. I cannot assess whether the analysis is actually possible (e.g., linking data from different registers using unique personal IDs) and whether these are in accordance with the respective regulations.
2. I publish all my reviews online at https://reviews-fb5c76.zivgitlabpages.uni-muenster.de.
3. I have very little experience with qualitative research and recommend to let a researcher with experience in qualitative research review study 3.
4. I used Claude Opus 4.8 and 5 for complementary assessment after I wrote the report. It raised the point of endogeneity and a tautological / unfalsifiable hypotheses, the latter of which is in line with the crud factor argument I raised. The former is specified above after I verified it against my own understanding.
5. I looked up if any of the cited studies were replicated or reproduced, which was not the case (via Replication Atlas)
6. I used Claude Opus 5 to check my review’s tone. I followed some of its suggestions to tone this critique down since I am aware that my criticism can be much more harsh than it needs to be.