thesis /
01-Chp-Background.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462---
bibliography: bibliography/references.bib
csl: bibliography/nature.csl
output:
bookdown::pdf_document2:
template: templates/brief_template.tex
bookdown::word_document2:
reference_docx: templates/word-styles-reference-01.docx
number_sections: false
bookdown::html_document2: default
documentclass: book
---
```{block type='savequote', include=knitr::is_latex_output(), quote_author='(ref:alzheimer-quote)', echo = TRUE}
Alles in allem genommen haben wir hier offenbar einen eigenartigen Krankheitsprozeß vor uns.
~
Considering everything, it seems we are dealing here with a special illness.
```
(ref:alzheimer-quote) --- Alois Alzheimer, 1907 [@alzheimer1907; @stelzmann1995]
```{r, echo = FALSE, warning=FALSE, message=FALSE}
source("R/helper.R")
knitr::read_chunk("R/01-Code-Background.R")
doc_type <- knitr::opts_knit$get('rmarkdown.pandoc.to') # Info on knitting format
```
# Background, Theoretical framework, Aims & Objectives {#background-heading}
\minitoc <!-- this will include a mini table of contents-->
\newpage
::: {.laybox data-latex=""}
## Lay summary {-}
Around 850,000 people in the UK live with dementia, and by 2040, nearly twice as many will have the condition. Despite many promising candidates, no cure for dementia currently exists, meaning the focus is on finding ways to prevent the condition. The best way to do this is to find risk factors (characteristics that influence a person's chance of developing a disease) for dementia that we can easily change. Avoiding a risk factor does not guarantee that a person will not develop dementia but makes it less likely. A key risk factor for dementia may be the levels of lipids (fatty substances such as cholesterol) in a person's blood, though not all existing research agrees. The aim of this thesis is to use all available evidence to assess whether [raised]{.correction} blood lipid levels are in fact a [cause of]{.correction} dementia.
This introductory chapter provides background information on both dementia and blood lipids, and on the potential link between them. It introduces the theory used to frame the research presented here, and then maps the formal aims and objectives of the research project to the relevant chapters of this thesis. Finally, it summarises the outputs (journal articles, presentations and software) that were created as part of this thesis.
:::
<!----------------------------------------------------------------------->
## Introduction
This chapter provides an overview of the broad context of this thesis, introducing the core concepts used throughout and providing some background on each. It briefly discusses the underlying pathologies and diagnosis of dementia, its public health importance, and the current state of treatment and prevention research. It then provides background on blood lipids and lipid-modifying treatments, and summarises the types of evidence used to examine the effect of these exposures on dementia outcomes.
It then introduces evidence synthesis as the key framework used to guide the research presented in the remaining chapters. Finally, it outlines the aims, objectives, and structure of this thesis, and briefly summarises the contributions to the scientific literature that arose from this research.
<!--- No nb here ---------------------------------------------------------------->
## Dementia
### Definition and underlying pathologies {#underlying-pathologies}
Defined by the Diagnostic and Statistical Manual of Mental Disorders as a "major neurocognitive disorder", dementia is a progressive disease which impairs cognitive functions including speech, memory and executive reasoning.[@edition2013] At advanced stage, the condition causes severe behavioral and personality changes,[@cerejeira2012] culminating in reduced motor control that affects patients' ability to swallow or breathe.[@kumar2013]
The condition is thought to have several distinct underlying causes, the most common of which are Alzheimer's disease and vascular dementia.[@burns2009] Other causes include rarer dementia subtypes (e.g., Lewy body or frontotemporal dementias) and other neurological diseases (e.g., Parkinson's disease).[@burns2009]
Alzheimer's disease is the most common cause of dementia, accounting for approximately 60-80% of [cases.]{.correction}[@alzheimersassociation2020] Characterised by substantial cognitive impairment and difficulty with high level executive function, it is an insidious disease, with initial onset thought to occur up to 15 years prior to symptomatic presentation.[@robinson2015] Much remains unknown about Alzheimer's pathogenesis, despite research implicating the "amyloid hypothesis",[@robinson2015] as a potential mechanism of disease. Under this hypothesis, the build-up of amlyoid plaques (composed mainly of amlyoid-$\beta$ peptide) and neurofibrillary tangles (composed mainly of tau protein) triggers a range of physiological changes, including inflammation and cell death, that result in cognitive impairment.[@robinson2015]
Vascular dementia (VaD) is the second largest underlying pathology of dementia, accounting for ~10-25% of [cases.]{.correction}[@alzheimersassociation2020] Vascular dementia is caused by a range of cerebrovascular disorders, and as a result, presentation of symptoms can vary widely.[@iadecola2013] Similarly, due to the varied underlying pathophysiology, vascular dementia can onset either quite rapidly following a cerebrovascular event such as a stroke or over a long time-frame due to a series of small infarcts.[@venkat2015]
[While widely used in the published literature, the validity of these dementia subtypes as distinct clinical outcomes is debatable.]{.correction} Vascular dementia regularly co-occurs in patients with Alzheimer's disease,[@iadecola2013] [a]{.correction} presentation described as "mixed" dementia.[@custodio2017] [Evidence from autopsies performed as part of longitudinal population based samples, such as the Medical Research Council Cognitive Function and Ageing Study,]{.correction}[@chadwick1992] [found that a majority of participants had evidence of both Alzheimer's disease and vascular pathology.]{.correction} [@neuropathologygroupofthemedicalresearchcouncilcognitivefunctionandageingstudymrccfas2001] [This suggests that the "pure" clinical subtypes of dementia are less common than the "mixed" outcome, particularly in older age groups,]{.correction}[@schneider2007] [and indicates that a composite "all-cause" dementia outcome may be more analytically valid than outcomes based on a single dementia subtype.]{.correction}
<!----------------------------------------------------------------------->
### Diagnostic criteria {#diagnostic-criteria}
Dementia is difficult to diagnose, primarily due to its slow onset, in addition to the confusion of initial symptoms with normal ageing.[@robinson2015] Dementia is diagnosed on the basis of behavioral and cognitive changes as assessed by an experienced clinician, using one of several diagnostic criteria.
One of the most commonly used sets of criteria for diagnosing dementia cases are the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria (Table \@ref(tab:diagnosticCriteria-table)).[@edition2013] Outlined in Table \@ref(tab:diagnosticCriteria-table), these form the broad definition of a dementia diagnosis, and are supported by a detailed patient history, evidence from carers and family members, and objective assessments of cognitive ability using neurocognitive tests.
Many cognitive assessment tools exist for the purpose of informing a diagnosis of dementia,[@sheehan2012] with two of the best known of these being the Mini Mental State Exam (MMSE) and Montreal Cognitive Assessment (MoCA) scale. The distinction between these memory scales and diagnostic criteria presented above should be noted. For example, the MMSE is used to provide evidence for part A of the criteria presented in \@ref(tab:diagnosticCriteria-table). Taken alone, it does not indicate the absence or presence of dementia, instead merely indicating cognitive impairment which could have another cause (for example, temporary delirium as a result of an infection or surgery).
<!----------------------------------------------------------------------------->
(ref:diagnosticCriteria-caption) __Overview of the DSM-5 criteria for dementia (major neurocognitive disorder)__ - These published criteria[@edition2013] are commonly used to define dementia cases. Criterion A captures the cognitive decline aspect of dementia, while Criteria B-D ensure that the impairment observed is sufficient to interfere with daily living and cannot explained by other causes. Note that all criteria (A-D) must be met.
(ref:diagnosticCriteria-scaption) Overview of the DSM-5 criteria for dementia (major neurocognitive disorder)
```{r diagnosticCriteria-table, message=FALSE, results="asis", echo = FALSE}
```
<!----------------------------------------------------------------------------->
Differentiating between the underlying causes of a dementia diagnosis is challenging but necessary, because whether the patient has Alzheimer's disease or vascular dementia will affect expected progression and potential treatment options available (see Section \@ref(intro-treatments)). Cause-specific criteria exist for the diagnosis of dementia sub-types. For example, the NINCDS-ADRDA criteria are commonly used to assess patients for Alzheimer's disease,[@dubois2007] while vascular dementia is diagnosed using the NINCDS-AIREN criteria.[@roman1993]
<!----------------------------------------------------------------------->
### Public health importance
Dementia is quickly becoming a critically important public health issue. Despite the age-specific incidence and prevalence of dementia remaining relatively constant over time,[@prince2016] an ageing population is set to create a dementia epidemic, particularly in westernised countries.[@flier2005] While approximately 525,000 patients have received a dementia diagnosis, the true number of people currently living with dementia in the UK is thought to be closer to 850,000, with this figure expected to double by 2040.[@baker2019] Globally, the prevalence of dementia is expected to reach 75 million by 2030.[@prince2016] Dementia is a leading cause of death in the UK,[@zotero-15757] [and its impact on mortality is likely underestimated, given that death certificates frequently do not accurately capture dementia cases.]{.correction}[@ahmad2021; @gao2018] [Further,]{.correction} it is one of the few leading causes of death without a proven treatment.
[Additionally, multimorbidity is common among people with dementia, with patients frequently having two or more other chronic conditions,]{.correction}[@poblador-plou2014] [including diabetes, depression and cerebrovascular disease.]{.correction}[@bauer2014] [Dementia makes the diagnoses and management of comordid conditions harder, as people living with dementia may not be in a position to seek primary care]{.correction}[@cooper2017] [or self-manage treatments.]{.correction}[@sinclair2000]
[In addition to the patient-centered harm described,]{.correction} dementia also has a substantial economic impact. In 2015, the estimated total cost of dementia in England was £24.2 billion. Health care costs alone were £3.8 billion, with the remainder being divided between unpaid care and social care costs.[@wittenberg2019] Globally, the cost of dementia care is expected to rise to $1tr by 2030.[@prince2014] As such, the urgent need to reduce the burden of dementia, both at the personal and system level, is clear.
<!----------------------------------------------------------------------->
### Treatments {#intro-treatments}
Developing treatments for dementia is regularly deemed to be one of the most challenging markets in the pharmaceutical world, with trials of seemingly promising therapeutics being regularly abandoned due to futility.[@cummings2020]. At present, there are no known curative treatments for dementia, regardless of the underlying cause, though several available therapeutics can help alleviate the symptoms of Alzheimer's disease.
The most common of these are acetylcholinesterase (ACE) inhibitors, which inhibit the degradation of the neurotransmitter acetylcholine by competitively binding the ACE enzyme. Acetylcholine plays a key role in controlling the cholingeric synapses, which are highly concentrated in regions of the brain (such as the neocortex) that control higher level brain functions such as memory and attention.[@hampel2018] Commonly prescribed ACE inhibitors include donepezil and galantamine.[@pariente2008] ACE inhibitors increase the availability of the neurotransmitter, and have shown clinical effect is easing the behavioural and memory-related symptoms of Alzheimer's disease.[@marucci2020] ACE inhibitors represent only a stop-gap treatment, treating the symptoms rather than the underlying pathology which may continue to progress.[@francis2010]
<!----------------------------------------------------------------------->
### Risk factors
Given the substantial burden that dementia represents and the absence of any curative therapies, as detailed in the above sections, the assessment of easily modifiable targets for their utility in the prevention of dementia should be prioritized.[@winblad2016] [Previously-examined risk factors for the condition include cardiovascular factors (e.g. hypertension), education, diabetes, and smoking.]{.correction}[@anstey2019] [Similarly,]{.correction} the benefits of a prevention-based approach based on addressing known risk factors for dementia are well-studied. Reducing the prevalence of the seven most important risk factors for dementia (obesity, hypertension, diabetes, smoking, physical inactivity, [depression,]{.correction} and low educational attainment) by 10-20% per decade is estimated to result in a reduction in dementia prevalence of 8-15% by 2050.[@norton2014potential]
In this context [of preventative treatment]{.correction}, lipid levels represent a [potential]{.correction} target, due to the ready availability of lipid-modifying treatments which could be repurposed.[@pushpakom2019] Given the existing uncertainty in the evidence [(as discussed in Section]{.correction} \@ref(evidence-association) [, below),]{.correction} determining whether variations in lipid levels are causative for dementia may prove critical in reducing the future burden of the condition.
This thesis will focus on blood lipids as the primary risk factor of interest. The next section provides an overview of blood lipid fractions and therapeutic interventions that modify them, while Section \@ref(evidence-association) provides an overview of the existing evidence for an association between lipids and dementia outcomes.
<!----------------------------------------------------------------------->
## Serum lipids
### Lipid fractions {#intro-lipid-fractions}
The blood lipid profile contains a range of component parts, or fractions. However, this thesis will only consider the two most important fractions, triglycerides (TG) and cholesterol, which are either absorbed from food (exogenous lipids) or produced internally (endogenous lipids).[@feingold2000]
Triglycerides are the simplest and most common type of lipids, used to store excess calories from food and to move energy around the body.[@laufs2020] In contrast, cholesterol is primarily used to create cell walls and certain sex hormones.[@zampelas2019] Lipids are not water soluble, and so cholesterol is transported through the blood stream in lipoprotein structures of varying densities. Low-density-lipoprotein cholesterol (LDL-c), commonly known as "bad" cholesterol, acts as an energy conveyor by transporting fat to cells. In contrast, high-density-lipoprotein cholesterol (HDL-c) transports cholesterol to the liver to be broken down and excreted.[@feingold2000]
In addition to the individual fractions, total serum cholesterol (TC) is a commonly-used summary measure to estimate the total amount of lipid present in the blood. The level of TC is derived from measurements of the individual HDL-c, LDL-c and TG levels using the Friedwald formula:[@friedewald1972]
\begin{equation}
TC \approx LDL + HDL + kTG
(\#eq:total-cholesterol-formula)
\end{equation}
where $k$ is 0.20 if measurements are in milligrams per decilitre (_mg/dl_) and 0.45 if measured in millimole per litre (_mmol/l_). Widely used ranges for the acceptable levels of different types of lipids are based on the National Cholesterol Education Program (NCEP)[@national2002third], and are outlined in Table \@ref(tab:lipidLevels-table).
Elevated LDL-c in the bloodstream, a condition also known as hypercholesterolemia or hyperlipidaemia,[@nelson2013] can lead to atherosclerosis,[@libby2019] the build-up of fatty deposits in the blood vessels. These deposits constrict blood flow and can lead to vascular complications. Alternatively, part of the deposit can detach from the artery walls, forming a clot that can lead to a heart attack or stroke.[@libby2019]
<!----------------------------------------------------------------------->
(ref:lipidLevels-caption) __Classification of blood lipid levels__ - Category cut-offs for different lipid fractions according to the National Cholesterol Education Program guidelines.[@national2002third]
(ref:lipidLevels-scaption) Classification of blood lipid levels
```{r lipidLevels-table, message=FALSE, results="asis", echo = FALSE}
```
<!----------------------------------------------------------------------->
### Statins {#intro-statins}
Statins are a commonly prescribed method of lipid regulation.[@collins2016] Statins inhibit the conversion of 3-hydroxy-3-methylglutaryl-coenzyme-A (HMG-CoA) into mevalonate, by competitively binding with HMG-CoA reductase (HMGCR). This conversion is a key rate-limiting step in the cholesterol biosynthesis pathway (see Figure \@ref(fig:statin-mechanisam)), enabling statins to effectively reduce the production of LDL cholesterol.
<!----------------------------------------------------------------------->
(ref:statin-mechanisam-cap) __Statin mechanism of action__ - Statins inhibit HMG-CoA reductase which controls the conversion of HMG-CoA to mevalonate. This conversion is the rate-limiting step in cholesterol biosynthesis, and so statin use can reduce the production of cholesterol.
(ref:statin-mechanisam-scap) Statin mechanism of action
```{r statin-mechanisam, echo = FALSE, results="asis",fig.pos = "H", fig.align = 'center', fig.cap='(ref:statin-mechanisam-cap)', out.width='50%', fig.scap='(ref:statin-mechanisam-scap)'}
knitr::include_graphics(file.path("figures/background/statinPath.png"))
```
Several statin treatments have been widely available for some time (see Table \@ref(tab:statinOverview-table)). Depending on the statin and dosage prescribed, the average reduction in LDL-c concentrations ranges from 15% with low-intensity regimen (e.g., pravastatin 5 mg/day) up to 60% with a high-intensity regimen (e.g., rosuvastatin 80 mg/day).[@collins2016; @law2003] Statins also vary with regard to their lipophilicity (the extent to which they are lipid soluble), affecting their localisation within the body, with hydrophilic statins being concentrated in the liver and lipophilic statins circulating more widely.[@schachter2005] This may create a divide in the off-target effects of statins with differing lipophilicity, particularly given the ability of lipophilic statins to permeate the blood brain barrier.[@sierra2011]
<!----------------------------------------------------------------------------->
(ref:statinOverview-caption) __Overview of common statins__ - There are several commonly prescribed statins used to treat hypercholesterolemia. Their approval date (US), properties and lipid-lowering effect are described here.
(ref:statinOverview-scaption) Overview of common statins
```{r statinOverview-table, message=FALSE, results="asis", echo = FALSE}
```
<!----------------------------------------------------------------------------->
### Other lipid regulating agents (LRA)
There are several other interventions that can be used to modify a person's lipid profile, each of which act via different mechanisms. However, in general, these treatments are either used as adjunct (additional) treatments to statin therapy or are used in situations where statins are contra-indicated or not tolerated.
<!-- TODO see https://www.heart.org/en/health-topics/cholesterol/prevention-and-treatment-of-high-cholesterol-hyperlipidemia/cholesterol-medications for information on the indications for treatment - could present as a table . If using table, put reference to it in the above par -->
The most commonly used non-statin therapeutic is ezetimibe,[@kosoglou2005] which prevents intestinal absorption of cholesterol. However, when used alone, it has a limited LDL-c lowering effect, leading to the creation of combined statin/ezetimibe therapies (both compounds contained in a single pill, as opposed to complimentary treatments).[@genest2006]
Fibrates provide a second example of non-statin therapy. They are used to treat hypertriglyceridemia by reducing production of triglyceride carrying compounds in the liver. They are commonly used in patients with mixed hyperlipidaemia if treatment with statins has failed to sufficiently control cholesterol levels.
Finally, PCSK9 inhibitors (or PCSK9i) are a relatively new treatment with strong lipid lowering effects, lauded as a potential alternative to statins.[@chaudhary2017] Their mechanism of action is to bind to and inhibit PCSK9, which breaks down LDL-c receptors on the surface of the liver, thus allowing more LDL-c to be internalised and broken down.
Other therapies targeting triglycerides exist, including nicotinic acids[@mckenney2004] and omega-3-fatty acids,[@skulas2019] but they are far less effective in LDL-c lowering than the therapies described above.
<!----------------------------------------------------------------------------->
(ref:lipidTreatments-caption) __Summary of available treatments for hyperlipidaemia.__
(ref:lipidTreatments-scaption) Summary of available treatments for hyperlipidaemia.
```{r lipidTreatments-table, message=FALSE, results="asis", echo = FALSE}
```
<!----------------------------------------------------------------------------->
<!----------------------------------------------------------------------------->
## Evidence for the association between blood lipids and dementia {#evidence-association}
This section provides an overview of the varying sources of evidence on the relationship between blood lipid levels and dementia risk.
<!----------------------------------------------------------------------->
### Basic science {#intro-basic-science}
A role for lipids in the aetiology of dementia is supported by both genetic linkage studies and functional cell biology studies. The generation of the amyloid plaques found in the brains of Alzheimer's patients is cholesterol dependent,[@burns2003; @mizuno1999] while the most established genetic risk factor for late-onset dementia, apolipoprotein E (_Apo_$\mathcal{E}$), is involved in cerebral cholesterol transport. Several other genes involved in cholesterol transport have also been found to be associated with increased AD susceptibility.[@beecham2014; @harold2009; @meng2007]
Despite these results, evidence from the diverse range of epidemiological studies on this topic has been inconclusive.
<!----------------------------------------------------------------------->
### Observational studies
By far the largest source of evidence on the relationship between blood lipids and dementia outcomes comes from observational designs. Several studies have examined the relationships between concentrations of serum lipids (total cholesterol (TC), LDL-c, HDL-c and triglycerides) and both Alzheimer's disease and vascular dementia. These studies have reported extremely varied results. In some, a high serum cholesterol concentration has been found to be associated with an increase in susceptibility to AD,[@kivipelto2002; @kivipelto2005; @schilling2017; @solomon2009; @whitmer2005] however others have shown no association,[@li2005; @mainous2005; @mielke2010; @tan2003] or a reduced susceptibility [@mielke2005; @reitz2004]. With regard to vascular dementia, decreased levels of HDL-c appear to be associated with increased risk [@reitz2004], while for LDL-c, studies have reported both positive and negative associations.[@reitz2004; @moroney1999]
<!-- IDEA May need summary of NRSI here -->
<!----------------------------------------------------------------------->
### Randomised controlled trials
In terms of the central research of this thesis, randomised controlled trials (RCTs) of statin therapy can be used to provide indirect evidence on the effect of reducing blood LDL-c levels on dementia risk. However, RCTs may be infeasible if the outcome of interest is one with a long prodromal period, such as dementia (see Section \@ref(underlying-pathologies)), as they would require extremely long and costly follow-up.[@ritchie2015] It is no surprise then that the two previous trials providing evidence on the effect of statins on dementia risk, identified by a recent Cochrane review,[@mcguinness2016] are in fact trials of statins for the prevention of coronary-related outcomes.
While being widely cited, these studies have major limitations that reduce their utility as a source of evidence on the effect of statin treatment in assessing the impact of lipid-lowering treatment on dementia risk. Firstly, the criteria used to determine a dementia outcome are prone to misclassification. One of the trials, the Prospective Study of Pravastatin in the Elderly (PROSPER) trial,[@trompet2010] reported not on dementia outcomes but on the change in cognitive scores. As alluded to in Section \@ref(diagnostic-criteria), a "change in score" alone is insufficient to diagnose a dementia outcome. The second trial, the Medical Research Council/British Health Foundation Protection Study,[@heartprotectionstudycollaborativegroup2002] found no effect of simvastatin on dementia (`r estimate(1, 0.61, 1.65)`), but did not report how the outcome was assessed/recorded within the trial.
Additionally, the two trials did not make any effort to assign an underlying pathology to each case, instead reporting an all-cause dementia outcome. As discussed in Section \@ref(underlying-pathologies), the different underlying pathologies of dementia have different mechanisms of action, and so it is not guaranteed that the effect of statins would be consistent across them.
Both trials were also limited by the relatively short follow-up period examined. This limited follow-up was to be expected because the primary aim of both trials was to assess the effect of statins on short-term coronary-related outcomes.[@heartprotectionstudycollaborativegroup2002; @trompet2010] The PROSPER trial had a mean follow-up of 3.2 years, while the MRC/BHF [Heart]{.correction} Protection Study estimated risk at 5 years of follow-up. Given the long lag time between non-symptomatic onset of dementia and clinical presentation, it is likely that these durations are insufficient to fully capture the onset of dementia. [Further]{.correction}, as they included only patients at high vascular risk, their generalisability to other settings is limited.[@mcguinness2016]
[A final general issue with studies of statin use is their limited ability to assess a causal link between lipid levels and dementia. This is due to the potential pleiotropic effects of statins, which may effect dementia risk via some non-lipid-lowering pathway. For example, previous work has shown that statins influence the levels of C-reactive protein,]{.correction}[@albert2001] [an independent risk factor for dementia.]{.correction}[@lewis2021; @schmidt2002]
<!----------------------------------------------------------------------->
### Mendelian randomisation {}
Newer methodological approaches, such as Mendelian randomisation (MR),[@daveysmith2014] have also been used to examine the effect of varying lipid levels on dementia risk. Mendelian randomisation attempts to combat the risk of reverse causation and residual confounding inherent to observational studies by using natural random variation in participants' genomes.[@greenland2000] In brief, MR uses genetic variants that are both strongly associated with the exposure of interest and are independent from potential confounders to strengthen causal inference.[@daveysmith2014] The analytic method relies on several assumptions about the instrumental variable (IV)[@davies2018] which are discussed in Section \@ref(rev-discussion-MR).
Recent MR studies indicated that genetically determined low levels of LDL-c may cause a reduction in AD risk.[@larsson2017; @ostergaard2015] However, the effect was attenuated in sensitivity analyses that exclude the region surrounding the _Apo_$\mathcal{E}$ gene, the strongest known risk factor for Alzheimer's disease.[@kim2009] Inclusion of _Apo_$\mathcal{E}$ variants invalidates the exclusion restriction criteria [of MR analyses]{.correction}, as the risk reduction observed may be driven by variants in this region via a pathway independent of lipid levels. This was supported by further MR studies in which the observed effect attenuated to the null after _Apo_$\mathcal{E}$ variants were excluded.[@benn2017] Despite the increasing number of MR studies examining this topic, no systematic review of this study design as a source of evidence has been performed.
In summary, multiple sources of evidence exist on the relationship between statins and dementia. In the next section, I introduce the synthesis of diverse sources of evidence as the theoretical framework used in this thesis.
<!----------------------------------------------------------------------->
## Theoretical framework: Evidence synthesis
Evidence synthesis is the process of finding and integrating information from several sources to examine a research question.[@donnelly2018] The results of an evidence synthesis exercise can be used to provide a more definitive answer to that question or, failing that, to highlight gaps in the existing evidence base. The ability to identify these gaps is particularly useful in guiding future research to address questions that have yet to be answered. A common type of evidence synthesis exercise is a systematic review, either with or without a meta-analysis.[@chandler2019chapter]
This thesis seeks to use an evidence synthesis framework to assess the effect of lipid levels, and treatments that influence lipid levels, on dementia outcomes. Specifically, this thesis considers three concepts within the umbrella term of evidence synthesis:
* Inclusion of preprints
* Individual participant data meta-analysis
* Triangulation across evidence sources
These three elements are expanded on below and are used to frame the research presented in the subsequent Chapters.
<!----------------------------------------------------------------------->
### Inclusion of preprints {#diverse-sources-preprints}
The importance of including grey literature, defined as literature not published in peer-reviewed journals,[@paez2017] in systematic reviews is widely acknowledged. Meta-research studies have demonstrated that systematic reviews excluding grey literature sources overestimate the effect of interventions.[@conn2003; @mcauley2000; @hopewell2007] Common, well-accepted forms of grey literature include conference abstracts, government reports and theses.[@paez2017;@lefebvre2019searching]
However, the role of preprints in evidence synthesis is less well defined. Described by the Committee on Publication Ethics as 'scholarly manuscript[s] posted by the author(s) in an openly accessible platform, usually before or in parallel with the peer review process',[@committeeonpublicationethicscope2018] preprints serve several purposes. They are used to establish primacy when submitting to a journal where the peer-review process may take several months.[@vale2016] In addition, [they]{.correction} allow for the rapid dissemination of research findings, as occurred during the COVID-19 pandemic.[@fraser2020preprinting] Finally, they are a source of information on manuscripts that may not have been accepted elsewhere, helping to combat publication bias or the "file-drawer" effect.[@rosenthal1979]
One of the major criticisms of using preprints as an evidence source is that they have not yet undergone formal peer review.[@maslove2018; @schalkwyk2020] However, this approach assigns substantial weight to peer-review as an indicator of "quality", and is at odds with the acceptance of non-reviewed conference proceedings as an evidence source.[@lefebvre2019searching; @mahood2014] The argument for including preprints as an evidence source is further strengthened by results that demonstrate preprinted studies seldom change following peer review. Meta-studies of the concordance between preprinted and published studies showed that results were broadly comparable between the two, indicating that while the numerical results may change, the overall interpretation of the results were consistent in the majority of cases.[@shi2021; @klein2019; @nicholson2021] This indicates that preprints should be considered a reliable reflection of a given study.
In this thesis, preprints are considered an important source of evidence, in contrast to previous reviews on this topic. However, as with grey literature,[@mahood2014] there are several logistical problems with carrying out systematic searches in preprint repositories. As such, to enable the inclusion of preprints in the systematic review described in Chapters \@ref(sys-rev-methods-heading) & \@ref(sys-rev-results-heading), a new tool addressing these issues is presented in Chapter \@ref(sys-rev-tools-heading).
<!----------------------------------------------------------------------->
### Individual participant data meta-analysis
Individual participant data (IPD) meta-analyses are commonly considered to be the gold standard in evidence synthesis methodology.[@riley2010; @stewart1993] IPD methods seek to obtain the raw data from each study identified in a systematic review, rather than basing the meta-analysis on summary results extracted from the literature.[@riley2010]
In the context of this thesis, if lipids are found to have a causal role in development of dementia, evidence-based preventative strategies would be best informed by identifying the types of individuals who are most likely to receive benefit from treatment with lipid-modifying agents.[@arain2009; @clayton2018; @mccartney2016] However, if primary studies do not present results stratified by covariates of interest, meta-analyses of summary-level data on this topic often have limited ability to examine research questions related to exposure-covariate interactions.[@riley2010] Based on existing evidence, participant age and sex are considered to be of particular interest.[@arain2009; @letenneur1999]
An IPD meta-analysis of lipid levels on dementia outcomes would overcome this limitation of summary-level data because access to the raw data allows for an analysis that investigates these interactions.[@riley2020] This approach has the added benefit of allowing a common set of inclusion criteria and a common statistical model to be applied across all datasets, potentially eliminating some important causes of heterogeneity.[@stewart2002]
Despite their advantages, IPD meta-analyses are rarely performed.[@tugwell2010] Factors limiting their uptake include the increased time and effort they require when compared to a summary-level analysis, and the low success rate associated with obtaining the raw data.[@nevitt2017; @ventresca2020] The data underlying primary studies are frequently not publicly available,[@alsheikh-ali2011; @federer2018] and the availability of data "available on request" from authors declines rapidly over time.[@vines2014] Several systematic barriers to open data sharing have been identified.[@vanpanhuis2014] Of particular concern for biomedical IPD analyses are legal issues surrounding the sharing of medical data, motivated by concerns around patient privacy.[@wartenberg2010]
In response to these limitations, new collaborative initiatives have developed to enable rapid access to relevant data in a secure supported workshop. The most import in relation to this thesis is the Dementia Platform UK (DPUK),[@bauermeister2020] which aims to provide access to several dementia-related datasets via a single simplified application process.
[As part of this research, I attempted]{.correction} to obtain the raw data from relevant primary studies identified by the systematic review presented in Chapters \@ref(sys-rev-methods-heading) & \@ref(sys-rev-results-heading). Data obtained [were]{.correction} combined with that available from the DPUK portal as part of an IPD meta-analysis in Chapter \@ref(ipd-heading). This [enabled]{.correction} the assessment of the effect of lipids on dementia, and whether this relationship is modified by key variables such as participants' age or sex.
<!----------------------------------------------------------------------->
### Triangulating across study designs {#intro-triangulation}
As illustrated in Section \@ref(evidence-association), several diverse epidemiological methods have been used to examine the effect of varying blood lipid levels on dementia risk. Aetiological triangulation is a developing evidence synthesis method that seeks to exploit these differences in study design, and as a result, in underlying bias structures.[@lawlor2016] If several sources of evidence are available and point towards identical conclusions about an exposure-outcome relationship, and these sources are at risk of unrelated biases, this strengthens our confidence in the result. The ideal scenario is where predicted sources of bias are likely to be in competing directions, one to strengthen the effect of the exposure and the other to attenuate it.[@lawlor2016] As such, triangulating these results can provide a middle ground between the competing directions of bias. A triangulation approach can also prove useful in helping to prospectively design new studies that are at risk of different sources of bias to those already observed in the published literature.[@munafo2018] However, existing methods for triangulation are limited to a qualitative discussion of the different identified results with respect to their biases,[@lawlor2016] an approach which faces issues in interpretability when assessing many results.
To address this limitation, this thesis seeks to develop and apply a novel quantitative triangulation approach to answer causal questions on the effect of blood lipids on dementia outcomes. All existing evidence, regardless of study design, is first identified and assessed for risk of bias by the systematic review presented in Chapters \@ref(sys-rev-methods-heading) & \@ref(sys-rev-results-heading). The existing evidence base is then supplemented by two primary analyses, presented in Chapters \@ref(cprd-analysis-heading) and \@ref(ipd-heading). Finally all existing and new evidence is incorporated into the quantitative triangulation framework in Chapter \@ref(tri-heading).
<!----------------------------------------------------------------------->
## Thesis overview {#background-thesis-overview}
### Aim and objectives
The central aim of this thesis was to infer the causal effect of blood lipid levels on dementia outcomes via evidence synthesis methods. The specific research objectives that this thesis seeks to address are:
* To create a tool that allows for the inclusion of health-related preprints in evidence syntheses in a systematic and reproducible manner
* To review all available evidence across multiple diverse study designs to assess the effect of lipids and lipid regulating agents on dementia risk
* To examine whether there is evidence for an effect of lipid-regulating agents on dementia and related outcomes in a large-scale population-based cohort, the Clinical Practice Research Datalink (CPRD)
* To produce evidence on lipid-covariate interactions as part of an IPD meta-analysis
* To propose a generalised framework for the quantitative triangulation across diverse evidence sources
<!----------------------------------------------------------------------->
### Structure {#thesis-structure}
Chapters are self-contained, presenting the methods and results of that specific research project. The sole exception to this is the systematic review, which due to its size is split across Chapters \@ref(sys-rev-methods-heading) (Methods) & \@ref(sys-rev-results-heading) (Results). Each chapter is bookended by introductory and summary sections which place the methods and results in context. Finally, each chapter is prefaced by a "lay" or plain English summary.
- **Chapter \@ref(background-heading):** Background information on dementia and blood lipid levels. This chapter provides an introduction to the topics covered in this thesis and discusses the motivation for the research presented in the following chapters.
- **Chapter \@ref(sys-rev-tools-heading):** This chapter introduces a new tool, `medrxivr`, which was developed to allow for systematic searches of the health-related preprint repositories.
- **Chapters \@ref(sys-rev-methods-heading) & \@ref(sys-rev-results-heading):** These chapters describe, respectively, the methods and results of a comprehensive systematic review and meta-analysis. This review examined all available evidence on the association of blood lipids and LRA with subsequent risk of dementia.
- **Chapter \@ref(cprd-analysis-heading):** This chapter examines the relationship between lipid-regulating agent use and dementia outcomes in the Clinical Practice Research Datalink, a large primary care electronic health record database.
- **Chapter \@ref(ipd-heading):** This chapter describes an individual participant data analysis of three longitudinal cohort studies to describe the relationship between blood lipids and dementia outcomes
- **Chapter \@ref(tri-heading)**: This chapter integrates the evidence identified and produced by the preceding chapters as part of a quantitative triangulation framework.
- **Chapter \@ref(discussion-heading)**: This chapter summarise the key clinical and methodological findings of the thesis and discusses their implications for practice. The overall strengths and weaknesses of this project are discussed in detail, and further avenues of research are suggested.
<!----------------------------------------------------------------------->
## Outputs from this thesis {#thesis-output}
The outputs of this thesis are detailed below, and include peer-reviewed papers, presentations, and open-source evidence synthesis tools.
<!----------------------------------------------------------------------->
### Contributions to the literature {#literature-contributions}
During the course of this thesis, I have made several contributions to the scientific literature. Those arising from or directly related to the contents of this submission are presented below.
_**McGuinness L. A.**, Higgins J. P. T., Walker, V. M., Davies, N. M., Martin, R. M., Coulthard, E., Davey-Smith, G., Kehoe, P. G., and Ben-Shlomo, Y. (2021) Association of lipid-regulating drugs with dementia and related conditions: an observational study of data from the Clinical Practice Research Datalink. medRxiv [10.1101/2021.10.21.21265131](https://doi.org/10.1101/2021.10.21.21265131)_
Preprinted manuscript of the analysis of lipid-regulating agents and dementia outcomes in the CPRD, which is presented in Chapter \@ref(cprd-analysis-heading).
_**McGuinness, L. A.**, and L Schmidt. (2020) medrxivr: Accessing and searching medRxiv and bioRxiv preprint data in R. Journal of Open Source Software 5.54 2651. DOI: [10.21105/joss.02651](https://doi.org/10.21105/joss.02651)_
A paper introducing the open-source preprint search tool described in Chapter \@ref(sys-rev-tools-heading). As is common for journal articles describing software, the paper is intentionally short providing only a broad overview of the tool while extensive documentation is available from the project website (see Section \@ref(sys-rev-tools-intro) for more details).
_Hennessy, E. A., Acabchuk, R., Arnold, P. A., Dunn, A. G., Foo, Y. Z., Johnson, B. T., Geange, S. R., Haddaway, N. R., Nakagawa, S., Mapanga, W., Mengersen, K., Page, M., Sánchez-Tójar, A. Welch, V., **and McGuinness L. A.** (2021). Ensuring Prevention Science Research is Synthesis-Ready for Immediate and Lasting Scientific Impact. Prevention Science . DOI: [10.1007/s11121-021-01279-8](https://doi.org/10.1007/s11121-021-01279-8)_
The experience of extracting data for the systematic review in Chapters \@ref(sys-rev-methods-heading) & \@ref(sys-rev-results-heading) inspired a practical guide which aims to help primary researchers make their data synthesis-ready. This piece was co-written with Dr. Emily Hennessy (see Acknowledgements in the front materials).
_**McGuinness, L. A.**, and Higgins J. P. T. (2020) "Risk‐of‐bias VISualization (robvis): An R package and Shiny web app for visualizing risk‐of‐bias assessments." Research Synthesis Method). DOI: [10.1002/jrsm.1411](https://doi.org/10.1002/jrsm.1411)_
A publication describing the tool used to visualise the risk-of-bias assessments in Chapters \@ref(sys-rev-methods-heading) & \@ref(sys-rev-results-heading) has been published in Research Synthesis Methods. See Appendix \@ref(appendix-robvis) for more details on this tool. Note that this publication does not describe the recently-developed functionality for producing bias direction plots, as described in Chapter \@ref(tri-heading).
_**McGuinness, L. A.**, and Sheppard A. L. 2020. “A Descriptive Analysis of the Data Availability Statements Accompanying Medrxiv Preprints and a Comparison with Their Published Counterparts.” PLOS ONE 16(5): e0250887. DOI: [10.1371/journal.pone.0250887](https://doi.org/10.1371/journal.pone.0250887)_
Using the tool described in Chapter \@ref(sys-rev-tools-heading), I led a "research-on-research" study to assess the concordance between the openness of data availability statements accompanying a sample of medRxiv preprints and their published counterparts.
For information on additional contributions to the scientific literature not directly related to this thesis, see Appendix \@ref(appendix-publications).
### Presentations/Talks
<!--- Include links here for slides for each talk/poster --->
_"Identifying and triangulating all available evidence on the effect of blood lipids and statins on dementia outcomes"_: Poster presentation, Alzheimer's Association International Conference 2021.
_"medrxivr: A new tool for searching for and retrieving records and PDFs from the medRxiv preprint repository"_: Accepted oral presentation abstract, Cochrane Colloquium 2020 (note: event was cancelled due to the COVID-19 pandemic).
_"On the shoulders of giants": advantages and challenges to building on established evidence synthesis packages, using the {robvis} package as a case study"_: Oral presentation, Evidence Synthesis and Meta-Analysis in R Conference (ESMARConf) 2021.
_"RoB 2.0: A revised tool to assess risk of bias in randomised trials"_: Webinar, co-presented with Dr. Theresa Moore as part of the Evidence Synthesis Ireland Methods Series.
<!----------------------------------------------------------------------->
### Software {#outputs-software}
**`medrxvir`**
An R package that allows users to search and retrieve bibliographic data from the medRxiv[@rawlinson2019] and bioRxiv[@sever2019] preprint repositories. See Chapter \@ref(sys-rev-tools-heading) for more details. Install a stable version of the package from the Comprehensive R Archive Network (CRAN), or alternatively install the development version from GitHub, using:
```{r, eval = FALSE}
# CRAN version
install.packages("medrxivr")
# Development version
devtools::install_github("ropensci/medrxivr")
```
**`triangulate`**
An R package containing functionality to implement the quantitative triangulation approach detailed in Chapter \@ref(tri-heading). At present, only the development version of the package is available, which can be installed from GitHub using:
``` {r, eval = FALSE}
# Development version
devtools::install_github("mcguinlu/triangulate")
```
**`robvis`**
An R package and associated `shiny` web application that allows users to visualize the results of the risk-of-bias assessments performed as part of a systematic review. See Appendix \@ref(appendix-robvis) for more details. Install a stable version of the package from CRAN, or alternatively install the development version from GitHub, using:
``` {r, eval = FALSE}
# CRAN version
install.packages("robvis")
# Development version
devtools::install_github("mcguinlu/robvis")
```
<!----------------------------------------------------------------------->
## Summary
* In this introductory chapter, I provided background information on the core elements of the central research question and framed the research presented in the context of a theoretical framework of evidence synthesis.
* I also outlined the central aim of this thesis and provided an overview of the individual analyses performed. I described the contributions of this thesis to the field, both in terms of peer-review publications and conference abstracts.
* Finally, I discussed the research tools and software developed to support the analyses performed in this thesis, one of which (the preprint search tool) is introduced more fully in the following chapter.