<<

Questions from Committee

1) Why did you perform 7:1 matching?

We extracted all available cases of dementia meeting our criteria. A 7:1 matching was used to maximise power, particularly for rare exposures, while designed to keep within the number patients that CPRD allow for a single study. We have added text to the methods section as follows:

“Seven controls were used to maximise power to detect associations with potentially rare exposures or covariates, while adhering to the limits of the data provider regarding maximum sample size for a single study.”

2) The effect size is marginal (upper bound of 95% CI is only 1.14) and the potential for residual confounding high (were these prescribed for disorders associated with dementia risk OR even for subtle symptoms of very early disease?

Strength of effect size.

The effect size referred to by the committee (OR=1.11 95% CI=1.08-1.14) reflects the use of ‘any’ ACB3 in the MEP vs none. We would not expect this particular effect to be large, because it averages the effect from those using less than 3 months ACB3 (where little or no effect is expected) with the smaller number using 3 months or more ACB3 (where a much larger effect is expected and indeed observed). The dose-response analysis also in table 2 is the more meaningful to judge the size of effect.

We have added a section (see response to reviewer 2 comment 35 translating the effect size for anti-depressants into number needed to harm to aid clinical interpretation).

Residual confounding

We acknowledge that residual confounding can never be completely excluded in observational research, particularly in pharmacoepidemiology, and much of the strength of our report rests in the range of analyses presented specifically to test different ways in which confounding might have affected our results. We have added an extra appendix to further explore this possibility (see final comment in response to reviewer 3) and we believe that these demonstrate that residual confounding is unlikely to account for the association between ACB3 and dementia. In particular:

• There is a strong dose-response relationship between exposure to ACB3 and dementia incidence (table 2). • There is little difference in the associations reported when we adjust for covariates measured before start or before the end of the MEP (table 3). • Several associations remain when we restrict only to those exposures occurring 15- 20 years before dementia diagnosis (table 4). • There is no independent effect of the main potential confounder (depression) in the multivariate analyses restricted to 15-20 years before dementia incidence or in the dataset overall (appendix 4). • If residual confounding by poorly recorded depression did account for our finding then we would expect to see less effect among those with severe depression recorded by their GP. Our new analysis shows that the effect of anticholinergic does not depend on the presence or severity of recorded depression during that time (appendix 4).

Conversely, these same analyses show that apparently significant association between ACB1 exposure and dementia incidence indeed be attributable to confounding. Here there is no dose- response effect, and the association between dementia and ACB1 medications does not persist when exposures more than 10 years before diagnosis are considered (table 4).

We have added the following to the result section: “Appendix 4 details a further analysis of the associations between depression, antidepressants, and dementia incidence. Depression diagnosis has no independent association with dementia incidence after adjusting for antidepressants and other covariates in the model. However antidepressants were consistently associated with dementia, this effect was not attenuated by controlling for depression, and there was no interaction between depression severity and anti-depressant use.”

3) You allow for a 4 year period to avoid protopathic bias but the “latent” period for dementia may be longer than that.)

We agree it is likely that the latent period for dementia before diagnosis exceeds four years in many cases. For this reason we conducted the analysis given in table 4 (stratifying exposure by time since index date), which essentially also tests lag times of 10 and 15 years.

This shows that the effects of ACB3 antidepressants and urological medications persist even for exposures 15-20 years before dementia diagnosis. For protopathic bias to account for this result urinary incontinence would need to be substantially increased 15-20 years before dementia diagnosis. We have acknowledged this possibility in our discussion section as follows:

“Lower urinary tract symptoms themselves have been linked to future dementia incidence (35) and may be a symptom of early neurodegeneration (36), but to account for our finding unrecorded urinary incontinence would need to be a substantial risk factor for dementia diagnosed between 15 and 20 years later.”

4) Some patients “without dementia” may be “incipient dementia or very early, as yet undiagnosed dementia and if we follow for long-enough they will have dementia”. This is another potential source of bias.

We agree, as is the case with any study of chronic disease with insidious onset (eg hypertension, diabetes etc) many of our non-dementia control group may in fact have dementia that is not yet diagnosed, or have incipient cognitive impairment not yet severe enough for a diagnosis.

Also, any misclassification would shrink apparent effect sizes toward null through regression dilution bias, although could not remove them completely or induce a negative correlation. An analysis restricting to controls surviving beyond the index date without dementia would introduce a positive bias associated with the requirement to survive beyond the index date, and in any case many patients with dementia are never diagnosed. We have added the following to the discussion section under ‘strengths and limitations’ “..Hence it is possible that some of our controls have undiagnosed dementia or early cognitive impairment. This misclassification would shrink estimated effects toward the null (OR=1) but would not remove them completely. “

5) In the database, what is the severity of symptoms required to make a diagnosis of dementia?

Primary care data only includes complete information on whether or not a patient has ever received a diagnosis of dementia, and the date on which this diagnosis was made. In the UK dementia diagnoses are most often made in secondary care by specialist memory services.

We have added the following text to the discussion under strengths and limitations:

“In the UK most dementia diagnoses are made by specialist memory services which are then communicated to primary care, and so our cases are very likely to represent genuine cases of dementia.”

Further we have found in a separate (complete but unpublished) study comparing primary care diagnoses with objective measures of dementia status that almost all diagnosed cases in English primary care meet objective criteria for dementia when compared against diagnosis equivalent to DSM-III-R. Hence dementia patient in the UK would need to meet a threshold equivalent DSM-III-R to receive a diagnosis.

6) Many patients have mixed dementia, often with vascular etiology. Some patients may have had cognitive decline not yet coded as dementia and then suffered a stroke.

We agree many patients will have mixed dementia. Clinical expression of dementia in an individual reflects contributions from different pathologies and so the effects of risk factors may well contribute to cognitive impairment and dementia irrespective of the ‘primary’ subtype. For this reason and the inconsistent way in which dementia subtypes are coded in primary care data we have not disaggregated our cases based on dementia subtype.

7) Not all editors were familiar with .

Response: Dosulepin is a (used to be known as Dothiepin) we have added this information under ‘frequency of use’ in the results section.

8) The committee wasn’t sure of novelty.

Novelty is addressed by the responses to comments 9,10 and 11.

In short, previous studies have suggested that ‘anticholinergics’ are associated with dementia incidence. However this finding is much too broad to be of practical use and previous studies have been too small and without enough follow-up or power to control for confounders. The uncertainty around this finding has meant that previous findings have been easily dismissed and have not been acted on. We considerably narrow the groups of drugs around which concern should remain and provide much stronger evidence on their effect on dementia incidence. 9) What this seems to add is more granularity in terms of anti-cholinergic burden and drug class, and we find that the answer is not straightforward and there’s not an overall class effect. Do we need studies on subgroups of anticholinergics?

First, ‘anticholinergics’ describes a very wide class of medications with different actions and effect on the brain. Previous studies have not disaggregated these effects by pharmacological action or indication.

Second, it has not been possible to rule out confounding with prodromal symptoms or risk factors of dementia as rigorously as we have in this study. This is because previous studies have had shorter follow-up periods, smaller samples preventing disaggregation or poor confounding control owing to incomplete data on covariates.

Finally, many drugs (used at any time by around 50% of the middle aged and older population) are currently considered to be ‘mildly’ anticholinergic, reflected here by our ACB1 category. As previous studies had summed these effects together with the better established potent anticholinergics it was not known whether and to what extent these mild anticholinergics posed a risk of dementia. A very common drug with a small increase in risk could have a potentially large impact on dementia occurrence.

This along with the failure to disaggregate classes has led to the creation of anticholinergic burden measures that simply sum medications. We have shown these to be inappropriate, at least for dementia risk.

10) This association is certainly not new, but the categorization of anticholinergics into 3 subgroups was new to us.

The breakdown into three subgroups (ACB1, ACB2, ACB3) is important. ACB1 represents a very prevalent group of drugs with ‘possible’ anticholinergic effects, whereas ACB3 covers the much smaller number of drugs that are definitely anticholinergic and have known short term cognitive effects. Previous studies had summed these together, it is important, particularly for users of ACB1 medications to understand whether previously observed associations are in fact driven by the ACB2/3 group or whether ACB1s are specifically implicated.

11) In which way will this change practice?

Up to now the evidence surrounding the effect of anticholinergics on future dementia incidence has not been clinically ‘actionable’ and has been easily discounted. The reasons for this are that high quality observational studies have been rare, and with the failure to disaggregate drug classes with very different indications and pharmacological actions it has not been possible to attribute harm to any group in particular, and hence to apply findings to any particular patient.

As a consequence of this, although clinicians recognise anticholinergics in their patients and the harms of ‘anticholinergics’ are known, no change in the prevalence of anticholinergic use has been documented in the past 20 years.

We are able to attribute many previously postulated effects either to the ‘lumping’ together of all anticholinergics, or to confounding with prodromes of incident dementia, while isolating the groups of drugs for which genuine concern with respect to increased dementia risk should still remain. Furthermore, our study identified that the time window for deprescribing these medications may be much earlier such as targeting middle age patients (15 to 20 years before dementia onset). That is, our study considerably clarified the target of potential harm. It gives clinicians clear messages which will help them make informed choices for the benefit of patients when they make shared decisions with patients regarding the choice of medications.

Clinicians should now focus on avoidance of anticholinergic antidepressants and urologicals in high risk groups, particularly where evidence for the effectiveness of these medicines is weak. Given our findings, the right clinician to consider deprescribing anticholinergic may be psychiatrists, neurologists, and urologists in addition to primary care clinicians treating unspecified symptoms such as sleep, depression, and urologic symptoms.

Conversely for gastrological or cardiovascular medications, these are not associated with increased dementia risk in our study despite previous concerns caused by their known or postulated anticholinergic effects.

More generally, demonstrating that the effect of anticholinergics (with respect to dementia incidence; one of their key postulated harms) varies widely by class calls into question the use of any scale that simply counts ‘anticholingeric load’ and that all of the postulated harms of anticholinergics need to be evidenced on at least a class-by-class basis estimated within clinical populations.

We have reorganised and strengthened the manuscript in several areas to highlight the novelty and impact on practice.

12) Could you consider including a box with different anticholinergic groups to help doctors choose the least harmful drugs?

Our study would not be able to reliably demonstrate the differences of effects of individual drugs within classes. Also, a lack of evidence for any particular drug being harmful through its under- representation in our dataset does not constitute evidence for its safety.

Nevertheless we have distilled our results into an accessible box classifying the strength of evidence of harm for the major drug classes that we have tested. (Box 1).

We agree that estimating the risks of individual drugs would be a useful further step. Our study is an important step towards this, though demonstrating the variation in risks associated with different classes, and does indicate the need for specific drugs to be evaluated in future prospective epidemiology studies, not necessarily with reference to their anticholinergic properties but within cohorts defined by disease indication, considering the full range of available treatments, and using methods appropriate to the identification of risk factors from a very large number of candidates.

13) Clinicians are probably aware that drugs with anticholinergic actions are more prone to produce side effects and people do talk about the anticholinergic burden of polypharmacy. This study lumps together some quite different drugs to look for a common effect on dementia risk. Is it really useful to put chlorpheniramine, procyclidine, and tolterodine together and posit a common toxic effect (increased dementia risk)?

Response: Existing anticholinergic research and practice has done exactly this; lump together drugs such as those listed above. A common toxic effect had been discussed extensively in previous literature. Our study explicitly tested this grouping and confirms that is inappropriate, hence results being given disaggregated by drug class. This is a major advance of our study.

14) Please improve the discussion of the limitations and discuss the biases and scope for confounding.

We have improved the limitations section, expanded on the discussion around biases for specific associations and added more detail of the analysis on depression as a potential confounder to a new appendix as described in responses to comments 1 and 48

15) Looks relatively tight and well conducted. A major issue that we are not sure it has been adequately reported (or analysed) is that individuals are likely to have multiple anticholinergic medications. How were these treated if: a) fall on the same class, b) fall in different classes.

If multiple medications from the same class are used then their exposure in terms of DDDs is added within each patient (and within each time period for table 4). Hence each DDD corresponds to the total exposure to anticholinergics within each class, irrespective of whether they are all from the same drug, different drugs used at different times during the MEP or different drugs use concurrently during the MEP.

Our main analyses are multiple conditional logistic regressions, with exposure within each class added as a separate explanatory variable. Hence the effect of each class is estimated taking into account the exposure to each other class as well as the confounding variables.

We have added to the manuscript sentence to explain this under ‘anticholinergic exposure’, and clarified the conditional logistic regression model under ‘statistical analysis’.

16) Please explain why the classes used make clinical sense. Certainly their effect appears to be very different (within classes) as that is the focus of the paper.

The anticholinergic classes we have used arise from the Anticholinergic Cognitive Burden (ACB) scale. ACB categorises all drugs based on the severity and the strength of evidence for their anticholinergic effect. We initially classified drugs in this way partly because

• Summing would prevent the effect of the ACB3 drugs (among which there is the greatest concern) being directly examined • To test the hypothesis that ACB1 and ACB2 drugs are also linked to dementia incidence • To enable calculation of the ACB score for a participant However as you say we hypothesised and tested for differences in effects within these groups. Drugs within ACB groups were classified a priori based on primary clinical indication using standard WHO ATC groupings. This correlates closely with their pharmacological action. Drugs within our classes have similar receptor bindings, compared to considerable differences between the classes. Hence both from a clinical and pharmacological point of view, and to address our main scientific hypotheses these groupings are sensible.

We have added a sentence to the methods section under ‘anticholinergic exposure’ “These categories separate the main indications for anticholinergic drugs and also correlate closely with their pharmacological action. Drugs within our classes have similar receptor bindings, compared to considerable differences between the classes.”

17) There is also a question of how accurate/adequate is the calculation of the DDD based on CPRD.

See response to comment 36 (reviewer 2).

Reviewer 1

18) Confounding by treated disorder is an acknowledged limitation. The authors refer to the possibility that as depression is a known risk factor for dementia, this may be confounding this relationship – I thought adding references to this evidence base would be helpful.

We have added references to the introduction on late-life depression being linked to dementia.

19) I note that the odds ratio for dementia risk is higher for ACB1 antidepressants (which would appear to be citalopram) versus ACB3 which would seem to run counter to the argument. It might be interesting to compare ACB1 and ACB3 with ACB0 antidepressants.

Indeed, close to the time of dementia diagnosis ACB1 anti-depressants appear to be more strongly linked to dementia than ACB3 anti-depressants. This may be attributable to residual confounding, a channelling effect whereby are avoided in those considered more cognitively impaired or frail, or a genuine effect on those who are closer to dementia incidence and so are more cognitively frail.

ACB1, ACB3 and ACB0 antidepressants were each included in our full multivariate model, with ACB0 antidepressants included as potential covariate. We have added further supplemental material showing the effect of ACB0 and depression from the multivariate models shown in tables e3 and 4. (see Appendix) These demonstrate that ACB0 antidepressants follow the same pattern of risk as ACB1 antidepressants, but that depression is not associated with dementia after adjustment for the use of antidepressant medication. We have added a reference to this analysis and a short explanation to the body of the main paper in the results section (see response to comment 48).

20) The strongest relationship amongst ACB3 drugs is for antiparkinsonian drugs – and the increased risk of dementia in people with Parkinson’s disease is well established.

Agreed, and we have now more explicitly discussed this in our discussion under ‘meaning of the study’. There was no association between non-anticholiergic antiparkinsonians (not shown, included in our model as a potential covariate) and dementia, and anticholinergics have been linked to AD pathology within cohorts of Parkinson’s disease patients.

21) The title would make more sense to me if the word pattern was changed to “level”

We have replaced ‘pattern’ with ‘timing and level’ to reflect both of the aspects we have tested.

22) The relevance of ACB score of 3 is not explained in abstract – could the authors refer to “most potent” ACs so the abstract is understandable without reading paper to the non-specialist

Our abstract now reads:

“14,453 (35%) cases and 86,403 (30%) controls were prescribed at least one definite (ACB score 3, ACB3) anticholinergic during the exposure period .”

23) (v minor point) sp dosulepin in discussion

We have corrected this.

Reviewer: 2

24) The first point relates to the positioning of the respiratory () results. The authors indicate that they do not have good capture of OTC use. When looking at the binary variable of respiratory medication use, the fact that an association wasn’t found could be because of lack of capture leading to few individuals with higher doses (assuming a dose-response relationship). In looking at etable2, page 44, there is a signal with the 365-1459 category, with vary few participants in the highest category. It seems that more study is required to say that we are in the clear with these agents.

This is a good point and we have amended the discussion to reflect this as follows:

“A small association between antihistamine use and dementia was observed that did not meet our threshold for statistical significance. Although very few patients (around 0.3% of the sample) are prescribed more than 365 DDDs of in our study, those with fewer prescriptions may also have a substantial OTC use of ACB3 antihistamines that are not recorded. Those who are prescribed more than 365 DDD may comprise a lower socio-economic status group among whom prescribed medications are free of charge hence preferred to OTC medication. Hence our results, although in need of independent confirmation particularly for antihistamines, should reassure these patients.“

25) Second, there is a lack of mention of the classes that had a lower association with dementia. How should we be interpreting these results? Gastro-intestinal ACB1 and ACB3 drugs, and cardiovascular ACB1 drugs fall into this category (apparent protective effect). There is no mechanistic hypothesis for why this gastro-intestinal drugs should have a negative association with dementia incidence, but we might expect the protective effect of cardiovascular medications with respect to dementia incidence to outweigh any possible harm. We have added to the discussion section as follows:

“We find a slight negative association between ACB1 and ACB3 gastrointenstinal drugs and dementia with an apparent dose-response effect, although there is no mechanistic explanation for this. ACB1 cardiovascular drugs were also negatively associated with dementia with prolonged exposure (OR=0.95, 95% CI: 0.91 to 99; etable 2), suggesting their protective effect outweighs any possible harm associated with their anticholinergic properties.”

26) Page 3, Line 18: please describe what is meant by “deprivation”.

We have added “level of deprivation of the area in which each practice is located ” to the abstract, and a reference to the Index of Multiple Deprivation in the body of the paper.

27) Page 3, results: For completeness sake, mention results for ACB3 antipsychotics.

We have added ‘antiparkinsonians’. Antipsychotic are inconsistently linked with dementia incidence in our results.

28) Page 6, selection of cases and controls. Please provide information in the methods about the validity of the dementia diagnosis codes in CPRD.

Our code list for dementia was derived by combining code lists of Imfeld et al with the Department for Health’s Quality Outcomes Framework (QOF) list of business codes for dementia. These two code lists overlapped almost completely. We then considered the combined list within our clinical team including a general practitioner, and included all of these in our final definition with the exception of those that related explicitly to ‘alcohol related dementia’. We believe these codes are specific for dementia, and sensitive with respect to diagnosed dementia because:

• Imfeld et al validated their code list by sending a questionnaire to a sample of GPs and discovered a PPV of 95% (in the cited paper and later personal communication) • It is unlikely that GPs would use a code not on the QOF list to record a diagnosis because their adherence to targets for dementia diagnosis rate are based on these codes. • There are very few individuals diagnosed by drug only, so we are confident that the vast majority of known and treated dementia cases are diagnosed as such. • In another study currently completed but not yet published we have linked English GP data from 100 practices to objectively determined diagnoses (using an algorithm validated against DSM-III-R diagnosis) from an epidemiologic study and have shown that there are almost no ‘false positive’ cases of dementia included in GP records. (See also response to comment 4). We have added the PPV to the discussion section and a sentence on how our code list was derived to the methods.

“Dementia diagnosis codes in CPRD reflect GP diagnoses well with positive predictive value of 95%”

“Dementia codes were derived by comparing a previous published list (23) with the Department for Health’s Quality Outcomes Framework (QOF) list of business codes for dementia. These two code lists overlapped almost completely. We then considered the combined list within our clinical team, all codes were included in our final definition with the exception of those that related explicitly to ‘alcohol related dementia’.”

29) Page 7, Line 34: Can we assume if cases and controls were matched according to UTS data history that the MEPs were the same? It would be helpful to make the link more directly.

Response: Yes, this is correct, we have added this test to the reference section:

“Matching on years of UTS data history ensures that the MEP is identical within sets of cases and controls.”

30) Page 8, line 17: it is not clear why all antihistamines would be scored as ACB1 as some are considered ACB3. Please clarify.

Only 6% of the antihistamine prescriptions recorded during our MEPs were not already rated on the ACB scale. The four most frequent of these (representing 87% of these prescriptions) were for non- sedating antihistamines , , Mizolastine, and . Non-sedating anti- are specifically designed to be less likely to pass the blood brain barrier (hence why they are advertised as non-sedating). As they are less likely to pass the BBB they are less likely to have central (cognitive) effects. Therefore according to the way in which ACB scores are determined they are all rated as score 1. Nevertheless there is considerable variation with respect to the placement of non-sedating antihistamines among different anticholinergic scales. Our sensitivity analysis using ADS instead of ACB did not find substantially different results.

31) Page 13, line 15: missing a “.”

Corrected, thanks.

32) mention the lower risk with GI medications. This should be mentioned in the discussion as well.

See also response to comment 25. We have added sentences to the end of the results section to describe these results.

33) It seems that some of the sentences that focus on interpretation of results in this paragraph may be best placed in the discussion.

We removed the phrases “ supporting a causal interpretation” and “suggesting residual confounding rather than a causal effect of ACB1 on dementia incidence” from the results section.

34) Based on the ORs, I wouldn’t consider these “substantial associations”

We have changed the word ‘substantial’ to ‘significant’.

35) Page 16, somewhere in discussion it would be helpful to put the magnitude of risk due to certain ACs into perspective for clinicians.

Thank you for this suggestion. We have calculated the number needed to harm over a typical follow-up period for a typical association and added this to the discussion under ‘implications’.

“While the associations reported here are moderate (odds ratios for different exposures between 1.1 and 1.3), given the high incidence of dementia they reflect a significant potential risk to patients. For example, the odds ratio for dementia associated with any use of ACB3 antidepressants 15-20 years before index date is 1.19 (95% CI: 1.1 to 1.29). A typical patient aged 65-70 years might normally expect a period incidence of dementia of around 10% over the next 15 years (35), so this odds ratio would be consistent with an absolute risk increase of 2% (1%-3%) over that period, corresponding to a number needed to harm of 50 (33-100).”

36) Page 17, line 42: It would be worth elaborating a bit more that while DDD is the best available method, it does not capture relative anticholinergic activity across classes and especially level of AC activity (especially ACB1: e.g. 1DDD for ACB1 is not equal to 1DDD for ACB3 with regard to AC activity)

Difficulty in comparing across classes is an important reason for us keeping group separate in our primary analysis. In any case we have made this point more clearly as follows:

“DDDs can be difficult to establish for certain medications yet represent the best available method for comparing the levels of exposure of different drug classes. Our findings did not change when we instead analysed the number of prescriptions to quantify exposure (results not shown). DDDs also do not capture the relative anticholinergic activity across classes. Commonly used scales including ACB scale typically assume that drugs 3 have three times the anticholinergic activity of those scoring 1, but this is difficult to justify. Hence we stratified our results by ACB1, ACB2 and ACB3 in primary analysis so as to not enforce any particular relative anticholinergic activity, and separated by classes within those groups to avoid directly comparing DDDs across groups.”

37) Page 20, line 27: ACB3 respiratory: Do the authors think they have enough information to make this claim given the poor capture of OTCs in this dataset? (see general comments above) It would be worthwhile mentioning this limitation briefly when mentioning the antihistamine results on page 17, Para 1.

We agree and have adapted the discussion as follows: “A small association between more than 365 DDD of ACB3 antihistamine use and dementia was observed that did not reach statistical significance. Although very few patients (around 0.3% of the sample) are prescribed more than 365 DDDs of antihistamines in our study, those with fewer prescriptions may also have a substantial OTC use of ACB3 antihistamines that are not recorded in our data. Those who are prescribed more than 365 DDD may comprise a lower socio-economic status group among whom prescribed medications are free of charge hence preferred to OTC medication. Hence this result, while potentially reassuring for regular users of antihistamines, requires independent confirmation“

38) Page 20, last para: the authors mentioned in the abstract that we should be moving away from measuring overall AC exposure and focusing on subgroups. I think this is an important point and it should be discussed under the methodological considerations for future research.

We agree, we have strengthened and moved this sentence to the methodological considerations for future research section, as well as adding to the ‘what this study adds’ box.

39) Covariates: how was maximum dementia severity and duration determined.

We believe that you are referring to maximum depression severity and duration in the covariates section. Severity was mainly determined by the Read code entered by the GP, with the highest ever level up to each time point taken as the severity. Duration was measured as the time between first mention of depression or symptoms of depression in the primary care records and the time at which covariates were assessed. This detail has now been added to the Appendix under ‘detail of covariates’

40) Question about other findings in eTable 2: worth commenting on why we might find significant associations with even lower DDD levels (antidepressants and urological agents)

If we discount the possibility that lower DDDs might be associated with dementia incidence, then it is possible that this reflects either residual confounding by indication or another unknown variable, or that the low recorded exposure reflects a larger exposure that is not captured in this dataset, if for example a patient joined from a different primary care practice or the medication was given in secondary care. We have added:

“In each case (ACB3 antidepressants, antiparkionsonians, and urologicals) a dose-response effect is seen with smaller but significant positive association between dementia and recorded use of less than 90 DDD (etable 2). It is possible that in some cases these low exposures reflect a longer exposure that is not captures in the patients current primary care record.”

41) Results are presented for “Other” for ACB3, but no drugs are presented in the drug table (etable1). What are the “other” meds for ACB3?

We have added the only drug in this group: methocarbamol representing 0.4% of ACB3 prescriptions to etable1

Reviewer: 3

42) Background:- This is clearly and concisely written. It would be relevant, in light of the main findings and possible confounding, to address the methodological problems raised by the potential link between antidepressant drug exposure and later dementia. There are substantial reports on links between neuropsychological deficits, dementias and depressive disorders. In some models, disadvantaged early social conditions may predispose to depression and, independently, to late onset dementias. These pathways and their effects at the level of large populations are not introduced here or in later accounts of methods. The authors may hold secure opinions on these topics but absence here and from later discussion is noteworthy.

This is an important point. We have added a discussion of the likely role of depression and depressive symptoms on our study, and additional analysis in the appendix to address this question.

43) Methods:- It is helpful in largescale epidemiological studies to compare disease incidence between the disorders of interest and other well-established ‘reference’ disorders. This becomes of greater relevance whenever disorders overlap (e.g. stroke preceding dementia, dementia following stroke) or of well-established association between a confounder of interest (e.g. stomach cancer and socioeconomic status). Demonstration would remove doubt that the research methods used here have not init provided expected disease incidences and/or expected strengths of association with confounders.

We have not investigated disease incidence, as we have a case-control study hence It is not possible to describe incidence of disease or modify the case disorder cannot be modified to stroke or another condition. We agree in longitudinal population based studies this is an important tool to ensure incidence estimation is accurate by comparing with another condition, but not used in case-control studies.

To describe our dataset we have shown the recorded prevalence of each potentially confounding condition in table 1. Estimates of univariable associations between dementia incidence and important indications from our data suggest moderate effects that are consistent with previous studies but would be unlikely to be able to induce the effects we have observed through residual confounding.

44) participants:- satisfactory description that allows complete reproducibility. The selection of cases and controls is based on an assumption that the date of GP recognition of dementia (or an equivalent) gives a reasonable estimate of date of dementia onset. This is contentious: current understanding of the development of dementia places onset of underlying neuropathology many years earlier than this. Preclinical phases are variously timed between 10 and 20 years before typical clinical diagnosis. If this is true – and periods longer than 6 years seem highly likely - then exposure to anticholinergics could have often occurred after onset of progressive neurodegenerative changes. In these circumstances, it would be unsafe to involve anticholinergics in the initiation of pathogenesis of dementia.

We acknowledge that dementia is often diagnosed late or not at all, as discussed in the response to comment 4. This is the main reason we conducted the analysis presented in table 4. Our findings regarding anticholinergic antidepressants and urologicals are robust when exposure is considered 15-20 years before index date. Although the possibility remains that the drugs are being used for incipient neurodegeration, the fact that there is no independent effect of the main indications during this time suggests this explanation is unlikely.

45) anticholinergic medication exposure:- this is well-done and, within the limitations of the available data, provides reliable estimates of total exposure. Probably unhelpful to use acronym DEP for drug exposure period when depression is a major confounder.

We have changed this to medication exposure period (MEP) throughout.

46) covariates:- Missing data are mentioned later but together unclear how much of other meds, SES, miscellaneous health variables are absent. Those covariates already implicated in increased dementia incidence (head injury, stroke, COPD, drug/alcohol abuse, social isolation/living group) are worthy of inclusion here.

SES is complete as it is supplied based on practice level postcode using census data. In common with all analysis of routine data all health conditions are regarded as present where a record exists and absent otherwise. We have clarified this limitation in the results section under ‘statistical analysis. “As with other analyses of routine data, diagnosed health conditions or other medications are assumed absent if there is no mention in the primary care record. Deprivation is based on practice location and so is known for all patients. “

47) Discussion:- This is well-structured. Possibly would benefit from some anticipation of scrutiny of the main findings in medico-legal settings.

We have been careful to ensure that all of our conclusions are fully supported by our data and those of previous studies, and in the context of the study strengths and limitations.

48) There remain concerns that a wider epidemiological perspective was not taken of the source database and the effect size of the major effects of depressive disorders on the proposed link between anticholinergic drugs and later dementias.

We have strengthened the introduction and the discussion with respect to the effect of depression on dementia incidence and its possible role as a confounder in some of the associations we have reported. We agree that depression is a potentially important confounder of the effect of antidepressants, particularly for the exposures close to the time of dementia incidence.

Although the association between depression and dementia incidence is well established, several recent studies have suggested that mid-life depression is not a strong risk factor for later dementia, most notably findings from the Whitehall II study, and this is further discussed in the recent Lancet Commision for Dementia Prevention, Intervention and Care. We have now directly examined the possibility of depression being a confounder for our observed associations as follows:

We believe our analysis in table 4 suggests that depression does not underlie our findings, but as pointed out above it is possible that mood changes occur up to 20 years before dementia diagnosis. So to address this further we have added an extra appendix with further details of the analyses presented in etable 3 table 4, specifically the showing the effects of depression (symptoms and severity of diagnosed depression) and of all antidepressant medication. In short we show that:

• While depression is associated with dementia incidence, this association does not persist at any time point once we adjust for the prescription of anti-depressant medication. • Collinearity is not a problem, while there is (as expected) a strong link between the diagnosis of depression and the use of antidepressants the correspondence is by no means perfect and does allow the independent effect of each to be estimated. • Adjusting for depression as recorded has almost no effect on the association between dementia and antidepressants, if observed depression is not a confounder, then is seems unlikely that unrecorded depression is a residual confounder. • If we include an interaction term between severity of diagnosed depression and anticholinergic use, our estimated association between use of ACB3 antidepressants and dementia incidence does not vary systematically across groups. If this result was due to poor recording of depression we might expect a stronger association to be seen for those with no diagnosis (and in whom antidepressant use might be a marker for unrecorded depression) and no association among those with severe depression.

While we acknowledge that it is impossible to completely rule out this explanation we believe that, taken together, the analyses presented suggest that it is unlikely.