Manuscript ID BMJ entitled "Physical Activity, Cognitive Decline and Risk of Dementia: a 28-year Follow-Up Study"

Transcription

1 BMJ - Decision on Manuscript ID BMJ Body: 07-Apr-2017 Dear Dr. Sabia Manuscript ID BMJ entitled "Physical Activity, Cognitive Decline and Risk of Dementia: a 28-year Follow-Up Study" Thank you for sending us your paper. We sent it for external peer review and discussed it at our manuscript committee meeting. We recognise its potential importance and relevance to general medical readers, but I am afraid that we have not yet been able to reach a final decision on it because several important aspects of the work still need clarifying. We hope very much that you will be willing and able to revise your paper as explained below in the report from the manuscript meeting, so that we will be in a better position to understand your study and decide whether the BMJ is the right journal for it. We are looking forward to reading the revised version and, we hope, reaching a decision. dr. Wim Weber European editor, The BMJ wweber@bmj.com *** PLEASE NOTE: This is a two-step process. After clicking on the link, you will be directed to a webpage to confirm. *** abc014de8b5f91a72e00e316b **Report from The BMJ s manuscript committee meeting** These comments are an attempt to summarise the

2 discussions at the manuscript meeting. They are not an exact transcript. Members of the committee were: Elizabeth Loder (Chair), Gary Collins (Statistics advisor), Sophie Cook, Kristina Fišter, John Fletcher, José Merino, Rubin Minhas, Georg Röggla, Amy Price, Tiago Villanueva, Wim Weber, Daoxin Yin. Decision: Put points Detailed comments from the meeting: We thought your study addresses an interesting and important research question. After discussion however, we felt that it did not add enough to take it any further towards publication in The BMJ. We had the following (minor) queries: Tables. Asterisks to denote p-values should be avoided. If p- values are to be presented then report exact ones, unless very small. First, please revise your paper to respond to all of the comments by the reviewers. Their reports are available at the end of this letter, below. In your response please provide, point by point, your replies to the comments made by the reviewers and the editors, explaining how you have dealt with them in the paper. Comments from Reviewers Reviewer: 1 Recommendation: Comments: Dear editor and authors, Thank you for the opportunity to review this manuscript which describes a complex and multi-component statistical approach to answering an important and topical public health issue. I feel that this paper provides a welcome addition to the emerging literature questioning the well-established

3 protective effect of physical activity on cognitive impairment / dementia. The longitudinal analysis using observational methods on a robust and large established cohort study serves two purposes: 1) to enable triangulation with other study designs, including the recent systematic reviews and experimental findings cited in the manuscript; 2) to enable in-depth exploration of the data including modelling trajectories, supporting alternative hypotheses for the conflicting literature to date. The paper describes four main analyses: 1) Investigation of whether a cross-sectional association between PA and cog decline changes with age (linear mixedeffect models): fully-adjusted models revealed some evidence (for total and mvpa) 2) Investigation of whether a longitudinal association between PA and cog decline differs according to level of physical activity (linear mixed-effect models): fully-adjusted models revealed no evidence 3) Investigation of whether a longitudinal association exists between PA and clinical dementia diagnosis (using Cox regression models): fully-adjusted models reveal no evidence 4) Investigation of whether there is an observable difference in trajectories of PA over time in those who ultimately has a dementia diagnosis and those who don't (using mixed models): strong evidence of difference, similar findings found with the sensitivity analysis case-control approach. I have a few additional comments, questions, and / or suggestions for the authors: 1) To me it seems that the manuscript would be enhanced by inclusion of a discussion of the poor validity of self-report physical activity measures, which is a major limitation of using observational data to corroborate or refute experimental data. The best self-report measures exhibit a correlation coefficient of when compared to objective PA measures [1] so studies using self-report measures of PA can actually only comment on whether one's reported PA is associated with the outcome, rather than the effects of the PA per se. In a standalone prospective observational study it is plausible that reporting error is likely to be non-differential between those who have the outcome of interest and those that don't so there is no risk of this introducing bias. However when an attempt is made to compare the findings with experimental evidence which have used objective

4 measures of PA, it is important to clarify that the two types of measures are not equivalent. 2) The outputs / units of measure presented in the tables are not clearly described and for a reader unfamiliar with this methodology will be unclear; they are referred to in the text as SDs but it would be helpful to have a more explicit description of how the estimates from the models should be interpreted. 3) The discussion section commences with a summary of the findings and states that the "cross-sectional association between PA and cognitive function was more than three times stronger at ages than at 50-60". It is not clear where this effect estimate is found so further explanation would be useful. 4) It is unclear why the partially-adjusted cross-sectional analyses are presented in Table 2, instead of the fullyadjusted equivalent results which can be found in a supplementary table. 5) I believe that Table 3 would benefit from additional text in the headers / footers describing the estimates / units (as per point 2) and to which test the p values refer to (interaction between level of physical activity) 6) The methods chosen to explore associations between PA and cognitive decline / dementia in this data set are varied and complex. The methods section may be enhanced for the generalist BMJ reader who may not have specialist epidemiological or statistical knowledge by inclusion of a clearer statement of the aims of the four models. 6) Please could the authors check reference 29 as it seems spurious in this context. Overall I think the paper is well-written and the findings of clear public health importance. With minor modifications and inclusion of a clearer description of the mixed-effect models with interpretation of the outputs that would be suitable for a generalist reader this will make an excellent addition to the literature. [1] Harris T et al. A comparison of questionnaire, accelerometer, and pedometer. Medicine & Science in Sports & Exercise. 2009; 41(7):1392

5 Additional Questions: Please enter your name: Dr Gemma Morgan Job Title: Clinical lecturer in public health Institution: University of Bristol Reimbursement for attending a symposium?: No A fee for speaking?: No A fee for organising education?: No Funds for research?: Yes Funds for a member of staff?: Yes Fees for consulting?: No Have you in the past five years been employed by an organisation that may in any way gain or lose financially from the publication of this paper?: No Do you hold any stocks or shares in an organisation that may in any way gain or lose financially from the publication of this paper?: No If you have any competing interests <A HREF=' (please see BMJ policy) </a>please declare them here: I am funded by NIHR as an Academic Clinical Lecturer in Public Health Medicine. I have previously published a paper reporting an investigation of the same hypothesis which has been cited by the authors in several places in this manuscript. I confirm I have received no support from any organisation for the submitted review; have no financial relationships with any organisations that might have an interest in the submitted review in the previous three years; and have no other relationships or activities that could appear to have influenced the submitted review.

6 Reviewer: 2 Recommendation: Comments: Minor revisions suggested This is a thorough and relevant examination into the effects of physical activity on cognition using a large sample size and considerable follow-up period. Because of the enormous burden of dementia, both on patients and on society, this is a very worthwhile topic to investigate. The authors describe that physical activity, in contrast to many previous examinations, has no effect on (change over time in) cognition. Furthermore, physical activity is shown to decline in the year previous to a dementia diagnosis, which is described as a possible reason for the positive associations in previous examinations. This paper could be a valuable asset to current literature without too many revisions. However, there are some additional analyses which could be interesting and may add to the certainty of the conclusion made. Furthermore, the discussion would benefit from some additional elaboration on why there were no effects found, especially focusing on the difference in sample characteristics compared to previous examinations. Abstract and Introduction: No comments Methods: - An interesting feature of this study would be to delineate effects of physical activity on specific types of dementia (i.e. Alzheimer s disease). It was briefly mentioned in the discussion that this analysis was not possible due to low sample sizes. However, with a large proportion of dementia cases accounted for by Alzheimer s disease it would be interesting to see whether effects of physical activity change when only these patients are included. - Cardio metabolic profile was measured as one of the outcomes. It would be interesting to examine the effects of this in the models because of the association between cardio vascular fitness and physical activity. - Effects on cognition were assessed only using the global cognitive scores, it would be interesting to replicate these

7 same analyses on cognition scores within domains as an absence of effect on one domain could mask effects on other domains. Discussion: - The rate of cognitive decline in this relatively young group would be expected to be less than in other examinations implementing samples of older individuals. This decreases the chance of detecting subtle effects, especially with many predictors included in the models, which may be elaborated on in the discussion. - Another factor which could be taken into account is the age of onset for the incidence dementia cases in the current sample, was this information available? Judging from the age at baseline and follow-up time many of the incident dementia cases likely have an age of onset dementia lower than 65 years old. These cases have a higher chance of being part of a group with higher genetic vulnerability (e.g. APOE e4 risk allele in Alzheimer s disease). This may have confounded results, since physical activity may not be as beneficial to genetically at-risk individuals (for Alzheimer s disease less effect in APOE e4 carriers has been described). Additional Questions: Please enter your name: Colin Groot Job Title: PhD candidate Institution: VU University medical center Reimbursement for attending a symposium?: No A fee for speaking?: No A fee for organising education?: No Funds for research?: No Funds for a member of staff?: No Fees for consulting?: No Have you in the past five years been employed by an organisation that may

8 in any way gain or lose financially from the publication of this paper?: No Do you hold any stocks or shares in an organisation that may in any way gain or lose financially from the publication of this paper?: No If you have any competing interests <A HREF=' (please see BMJ policy) </a>please declare them here: None Reviewer: 3 Recommendation: Comments: This is a thorough study that uses a renowned data set to investigate an issue of great interest, and it presents remarkable results. The authors elegantly show that physical activity levels go down up to 9 years prior to a diagnosis, which presents a strong argument against studies using a follow-up period shorter than that. This is a very important observation. The study s conclusion is that there is no effect whatsoever of physical activity in middle age on the risk of cognitive decline and dementia later in life. This is contrary to many previous studies, and also to a systematic review and metaanalysis that our group performed, in which we cautiously concluded after reviewing the collective evidence in view of the Bradford Hill criteria for causality, physical activity is likely to prevent dementia. Of course this is what most of us would like to believe, and we must be cautious of our own biased attitude. Nevertheless, I believe there are reasons to think the present study may not present the correct answer. As the authors mention, the methods used for the ascertainment of diagnosis are rather weak. This would bias the results if cases among the less physically active were missed more frequently than those among more active participants. This is possible, as Table 1 shows that a low level of physical activity was associated with being nonwhite, having no university degree, low occupational

9 position, not being married or cohabiting, smoking, heavy alcohol consumption, poor diet, a higher BMI, diabetes, hypertension, CVD medication and lower physical fitness. On the one hand, these health problems would make this group more likely to be in frequent contact with the health system and thus more likely to be diagnosed with dementia (rather than less), but on the other, this group is on average less likely to make optimal use of that system and more likely to receive less-than-optimal care (reference needed), dementia might be masked by the symptoms of other conditions (such as those related to alcohol abuse), and the concern for possible dementia might be pushed to the background by more immediate health issues. It s impossible to say how that would balance out, especially after adjustment for confounders, but it could influence (bias) the findings of the present study. This is further supported by reference 18 (Tolppanen et al 2015), who similarly found no effect of (leisure time) physical activity on dementia risk in the full study, but did find it in a subsection that was properly investigated for cognitive function. The authors reported that CAIDE participants were 1.21 times more likely to develop dementia during follow-up (95% confidence interval [CI], , adjusted for age and sex). It could be that this is a selection effect as the authors of the present paper posit, but the alternative is that Tolppanen et al s statement would have been more accurately put as CAIDE participants were 1.21 times more likely to *be diagnosed* with dementia during follow-up (95% confidence interval [CI], , adjusted for age and sex). In other words, up to 30% of cases in the open population might not have been diagnosed. I do not understand the authors assertion that bias from dementia cases misclassified as non-cases is unlikely to unduly affect our results because even with a 50% error rate leading to 329 misclassified noncases the proportion of correctly classified participants with no dementia would be almost 20 times higher. The analysis focuses on the risk of dementia, not on the risk of not receiving that diagnosis. In the extreme case that the missed cases were all among the less active half of the group, this would change a relative risk of 1.00 to around 0.67 for the active group compared to the inactive half (see Table 4, bottom rows). The measurement of the exposure variable, physical activity, seems decent. Objective measurement would have been

10 preferred, but one can hardly expect that and a long follow up period. Nevertheless some regression dilution bias is possible (though of course this would have no effect where the relative risk is 1.00). But there was a change in the questionnaire; from the wave onwards participation in 20 specific types of physical activity were specifically asked. This could have aided participants memories, which could have led to higher reported activity levels. Was there any change in the number of hours per week reported (or other PA parameter) apart from the expected reduction over time? Figures 1 and S3 suggest that this may have been the case, as they show that people generally became more active as they aged, though this was all due to mild physical activity. This is unexpected, in view of Jim Sallis statement that The decline in physical activity with age may be the most consistent finding in physical activity epidemiology (Med Sci Sports Exerc. 2000;32: ). However, if the exposure did indeed shift upward mid-way the observation period, it is not clear to me how this might have influenced the results. One further reason for scepticism is that a substantial body of evidence supports the notion of at least part of what we call dementia is related to vascular disease. The mere fact that vascular dementia is the second most common form of the condition, in combination with the generally accepted links between physical activity and vascular risk indicators and -disease makes it highly surprising that physical activity would not reduce dementia risk. To my mind, this would put a lot of established knowledge into question, and in view of the above mentioned limitations of this study, I provisionally hold on the notion that probably, physical activity does have a modest protective effect against dementia. However this is not an argument that the authors can address in this study. Minor issues Table 1: Values for mean number of hours of moderate to vigorous activity for the groups following recommendations (or not) are switched. Additional Questions: Please enter your name: Lennert Veerman Job Title: Senior Health Economist

11 Institution: Cancer Council NSW, Australia Reimbursement for attending a symposium?: No A fee for speaking?: No A fee for organising education?: No Funds for research?: No Funds for a member of staff?: No Fees for consulting?: No Have you in the past five years been employed by an organisation that may in any way gain or lose financially from the publication of this paper?: No Do you hold any stocks or shares in an organisation that may in any way gain or lose financially from the publication of this paper?: No If you have any competing interests <A HREF=' (please see BMJ policy) </a>please declare them here: I have no competing interests. Reviewer: 4 Recommendation: Comments: Dr. Wim Weber Editor, BMJ Manuscript ID: BMJ , Title: Physical Activity, Cognitive Decline and Risk of Dementia: a 28-year Follow-Up Study

12 This is a timely manuscript, emphasizing a research focus (the impact of physical activity on cognitive function and risk) that is gaining importance across various fields from Medicine to Neuropsychology and Neuroscience. The authors employ an impressive sample, replete with multiple assessments of both physical activity and cognitive function. The originality and importance of this manuscript are impressive strengths. The sample size and up to 28-year longitudinal design represent definitive advantages for addressing whether physical activity serves a protective role indeed, as the authors emphasize, the length of follow-up is critical given emerging evidence suggesting that the prodromal phase of some dementias (Alzheimer s Disease in particular) may by 10 years or longer. The primary research questions were clear, as were the corresponding analyses that the authors employed to address these questions. On the basis of their analyses, the authors conclude that physical activity in midlife does not reduce dementia risk nor was it related to cognitive decline, and that declines in physical activity between the dementia and control groups were not significantly different until approximately 9 years before diagnosis. The authors conclude that there was no evidence of a neuroprotective effect for physical activity, and that previous reports of physical-activity links to cognitive decline and dementia may be due to a reverse causation bias. On balance, I believe this paper will be of particular interest to the readership of BMJ. Despite my enthusiasm for the research question, longitudinal design, and its implications for further informing the burgeoning literature on physical activity and cognition, I would like to respectfully raise the following questions and/or concerns with the authors: Relevant literature The authors do an admirable job of citing recent empirical findings and reviews that fail to document clear links between physical activity and cognitive decline or dementia risk. To be sure, however, there have been long-term investigations that have demonstrated clear links. As such, it might be informative to cite findings such as the study by Erickson and colleagues (2011) that clearly linked physical activity to structural and functional brain changes, as well as diminished MCI/dementia risk, following a 9-year follow-up. Notwithstanding, I appreciate the greater weight carried by the meta-analytic reports.

13 Measures Perhaps one of my greatest concerns pertains to the concatenation of the 3 cognitive measures (memory, executive function, and language) to yield a global cognitive score, as described on page 7. The first concern is a minor quibble for any of these measures, were alternate forms employed at future longitudinal follow-ups (particularly for the memory measure)? A number of longitudinal studies of cognitive aging have clearly demonstrated the influence of generalized practice effects for retest intervals as long as 5 years (and beyond). In the present case, could generalized practice have influenced the trajectories of long-term cognitive change? Please clarify whether the same measures were administered at each follow-up assessment. The creation of the global cognitive score itself represents a critical concern in my opinion. While it is true that the creation of cognitive constructs (e.g., by employing structural equation modeling, SEM) can help dissattenuate for measurement error, this was not done in the present case (a unit-weighted approach was adopted, described on page 7). The issue of SEM vs. the unit-weighted approach to aggregate indicators aside, the bigger concern pertains to what this newly created global cognitive score reflects? While memory, reasoning, and language abilities certainly share associations, these markers also reflect very distinct cognitive constructs -- constructs that might and have shown (a) distinct trajectories of change over time, (b) different points of onset (inflection points) of cognitive decline (see Laukka et al., 2012, JINS), and (c) perhaps most notably, differentiable associations with physical activity (e.g., a number of publications and meta-analyses have demonstrated a stronger influence of physical activity on executive function than language, for example). By collapsing across these 3 distinct cognitive measures, my concern is that any potential differences (described above) may have been obfuscated in the global score. Is there no association between physical activity and cognition, or is the association restricted to certain cognitive domains (e.g., executive function)? Based on these concerns, my suggestion would be that the authors recompute the analyses for each of the 3 cognitive measures (to avoid masking potential domain-specific effects). Sample The dementia cases identified in the present study reflect allcause dementia comprised of many distinct subgroups.

14 Although I appreciate the authors points raised on page 15 (that the number of cases were too few to facilitate more nuanced analyses), my view is that some additional detail is required. For example, some basic descriptive information on the dementia subtypes would be valuable. One concern is whether the heterogeneity in subtypes of dementia cases influenced the reported pattern of results? For example, would risk of Alzheimer s or vascular dementia be influenced to a greater degree by physical activity? Some theorists clearly make this argument. Further, in direct keeping with the authors statement about the length of the prodromal phase and its potential impact on the findings, our research group published a paper in 2012 (Laukka et al., JINS) that explicitly demonstrated a clear difference in longitudinal change-point models of cognitive decline between AD and vascular cases (this article employed a time to diagnosis parameterization what is referred to as backward timescale in the present manuscript). As expected, the prodromal period for AD was much longer than VAD, and when decline did emerge for VAD in proximity to diagnosis, the magnitude of cognitive decline was accelerated relative to AD. Just as the trajectories of cognitive decline vary for dementia subtypes (cf Laukka et al., 2012, JINS), it strikes me as plausible that PA-cognition links could also be distinct for dementia subgroups. Again, I appreciate that the overall dementia sample was small, but my questions concern whether (a) the number of dementia cases (and subtypes) is in keeping with population prevalence estimates, and (b) can the analyses be computed for AD cases only (the largest subsample)? At the very least, this concern could be noted in the discussion as either a limitation or direction for further exploration. Design For the population in general, and the backward timescale (time to diagnosis) models in particular, how did the authors estimate the date (incidence) of dementia onset? Some detail here would assist the reader in evaluating the precision of the models. Statistical Models I deeply appreciated the authors emphasis on the need to employ longitudinal designs that span many years in order to address questions such as the present one. As the authors argue, such analyses are essential for ruling out reverse

15 causal relations between physical activity and dementia risk. However, the models employed for the present paper are not quite complete, particularly those incorporating estimates of change. Among the models estimated, the authors employed: i.) linear mixed models (page 10) to examine cross-sectional differences in baseline physical activity as a predictor of subsequent 15-year cognitive decline ii.) cox models (pages 10-11) to examine cross-sectional differences in baseline physical activity as a predictor of subsequent dementia risk over a 27-year follow-up To the extent that change is a central focus in this research, the models outlined in i.) and ii.) above could be further augmented to directly incorporate change as a predictor. For example, the linear mixed models could move beyond examining individual differences in physical activity at baseline as a predictor of cognitive change to explicitly modeling change in physical activity in relation to change in cognitive function. The reported analyses examine betweenperson differences in physical activity in relation to cognitive change. In my view, the relative import of this study could be increased considerably if the analyses were to contrast between-person effects (as typically reported in the physical activity literature) with documented within-person associations. It has long been known that analyses executed at separate levels of nested-data hierarchies (i.e., the between- and within-person levels) do not necessarily yield equivalent results. Referred to as the ecological fallacy (Robinson, 1950), results at the individual level may be of a particular magnitude and direction, but when aggregated at the group level, can not only differ in pattern but may also be influenced by between-person confounds (e.g., age, cognitive status) that may obscure the effect of interest. For example, it is both conceptually and statistically possible for predictors to account for large proportions of variance within-persons, but to exhibit relatively little or no effect when pooled across individuals (e.g., Hofer & Sliwinski, 2001). Similarly, research on ergodicity by Peter Molenaar (e.g., Molenaar, 2004) underscores the importance of examining within-person associations. In the present study, the authors could explicitly model change in physical activity with change in cognition (i.e., a correlated-slopes model), or employ time-varying covariation models (in the linear mixed framework) to examine the within-persons association of physical activity and cognition.

16 The time-varying models would require the use of personmean centering (see Hoffman & Stawski, 2009), where individuals effectively serve as their own baselines and estimates of within-person associations are not conflated by between-person (e.g., population mean) confounds. Similarly, although the Cox models were very informative, they also incorporate cross-sectional differences in baseline physical activity as a predictor of subsequent dementia risk over a 27-year follow-up. Again, employing change in physical activity would provide the more compelling test of their hypothesis. The use of joint growth-survival models (e.g,. McArdle et al., 2005) would include both baseline level and long-term change estimates of physical activity as predictors of dementia risk. Such models facilitate a joint estimation of survival (e.g., time to dementia diagnosis) and change (growth curves in physical activity) components of the longitudinal data, and incorporate both the intercepts and slopes as predictors of dementia risk. If the authors choose not to adopt such models, my suggestion would be that they temper their conclusions arguably the most robust test of their research initiative requires (a) examining change in physical activity with change in cognitive function (or dementia risk), and (b) demonstrating the absence (or presence) of the physical activity-cognition associations at the within-person level. Minor Points Page 4 provide a more explicit characterization of short follow-up (e.g., in terms of years) Pages 9-10 for the statistical analysis section, it would be most useful if the authors would also include the equations (at least as supplemental material). With my sincerest appreciation on a most interesting read, Stuart MacDonald Additional Questions: Please enter your name: Stuart MacDonald Job Title: Associate Professor Institution: University of Victoria

17 Reimbursement for attending a symposium?: No A fee for speaking?: No A fee for organising education?: No Funds for research?: No Funds for a member of staff?: No Fees for consulting?: No Have you in the past five years been employed by an organisation that may in any way gain or lose financially from the publication of this paper?: No Do you hold any stocks or shares in an organisation that may in any way gain or lose financially from the publication of this paper?: No If you have any competing interests <A HREF=' (please see BMJ policy) </a>please declare them here: