METHODOLOGICAL CONSIDERATIONS

The association between HT use and late-life dementia and cognitive status is complex and confounded by multiple psychological, medical, and behavioral variables in a woman’s life which act as predictors of HT use including race, financial status, education, obesity, diabetes mellitus, intake of supplements, and overall health status (Gleason et al. 2012). In other words, where one theory revolves around the time of initiation of HT, the other proposes that it might be the health status of the user rather than simply her age which explains the effects of HT on brain functions (Hogervorst 2013).

The observational studies in this thesis consist of two case control studies and two cohort studies. The greatest strengths in the present thesis are that almost all studies had a large sample size and long follow-up times. However, our follow-up time was still not long enough to investigate the association between premenopausal use of HT with AD. Similarly, we could not explore the association between induced menopause commencing before the onset of natural menopause with AD. This is due to the fact that in all four studies of the present thesis, the women were mainly postmenopausal.

The use of HT in three studies was ascertained from the Finnish national prescription register (study 1, 2, and 4), which is maintained by the Social Insurance Institution of Finland. Nationwide data on prescriptions has been collected by the Social Insurance Institution since 1995 with almost universal coverage (97% of all prescribed medicines).

Duration of medication (HT) use is calculated based on prescription purchase data. ATC codes for each purchase history are processed for each individual. Purchase history is used to calculate defined daily dose which is then used to determine the exposure period for each drug. The prescription register does not cover drugs used in public nursing homes or during stay in hospitals (Tolppanen et al. 2016). This may introduce a serious limitation to our studies, where AD is the main outcome (study 1, 2, and 4) since in Finland around 30% of persons with AD live in nursing homes or are institutionalized. However, this is unlikely to have affected the results in studies 1 and 4 because these studies included only those women who were community-dwelling at baseline. In addition, exclusion of exposure data 0-5 years before the outcome did not impact on the results and conclusions in these studies. Study 2 was also restricted to women who were community-dwelling and free of AD and dementia at baseline.

An important point to consider about register-based HT use is thatregister-based data would misclassify those who discontinued HT use before 1995 as nonusers. It is unlikely that this misclassification was differential in those who later developed AD and those who did not. Therefore it would lead to an underestimation (i.e. wider confidence intervals) of the association between HT use and AD with the register-based data (study 2).

The AD diagnosis in studies 1, 2, and 4 was obtained from the Finnish special reimbursement register, which has been determined previously to possess high validity and predictive value for AD diagnosis, thus eliminating the possibility of misclassification of AD (Solomon et al., 2014b). Diagnosis of probable AD in Finland is based on the DSM-IV criteria for AD and NINCS-ADRDA criteria and supported by abnormal MRI and/or CSF findings typical for AD (McKhann et al., 1984, American Psychiatric Association 1994). The Finnish Current Care Guideline recommends that all persons with AD are treated with anti-dementia drugs unless there is a specific contraindication (such as gastric ulcer/intestinal tract operation <6 months ago or severe asthma or chronic obstructive pulmonary disease which are contra-indications for acetylcholinesterase inhibitors). Though the special reimbursement register has high positive predictive value for AD diagnosis, we cannot rule

out the false negatives, because it may be that some women with AD did not receive a special reimbursement. This group of women can contaminate the control group. Though it would dilute our findings towards null, it is a point worth considering. For study 2, AD diagnosis was taken from the special reimbursement register from 1999-2009. We could not ascertain the AD diagnosis prevailing among our study population from 1989-1999, as the special reimbursement register for AD was only established in 1999 when acetylcholinesterase inhibitors became available in Finland. It is possible that women who had deceased before 1999, had preclinical AD or AD. However, the number of these women was small (n=619) and thus it is not likely to have exerted any major impact on the results.

Data on oophorectomy, hysterectomy, and radical hysterectomy and malignancy of cervix uteri, corpus uteri, uterus or ovary (ICD-10 codes C53-C56 and C57.0 and corresponding ICD-9 and ICD-8 codes) for study 1, and 4, was collected from the National Hospital Discharge Register. Registration of hospitalizations and prescriptions is mandated by law in Finland, but currently there are no validation studies on oophorectomy and hysterectomy codes in the hospital discharge register.

One major limitation in our studies is lack of ascertainment for the start of HT in relation to the onset of menopause as women were mainly postmenopausal in all studies. Self-reported use of HT in study 2 and 3 is subject to a recall bias, though this was unlikely considering the previous validation study conducted in the same cohort where a postal inquiry was shown to be a reliable method of recording long-term HT use in Finnish postmenopausal women [Sandini et al., 2008]. Moreover, self-reported use of HT was balanced in study 2 by validation of HT from registers for a slightly shorter duration of time than self-reported use.

The similar results obtained with both modes of data collection added reliability to the self-reported results. Another bias to be considered in cohort studies (study 2 and 3) is the selection bias and healthy user bias. We cannot rule out that women responding to postal questionnaires in study 2 and self-reported HT use in study 3 were more educated, with a healthy life style, and were more socially active than women not responding to these questionnaires.

Though register based studies effectively account for drugs dispensed at pharmacies, they cannot evaluate the actual use of those medicines at the individual level. A common limitation of observational studies is the non-random allocation of exposure, meaning that differences between HT users and nonusers may partially explain our findings. As is common with register-based studies, we could not account for important factors affecting the HT-AD association either acting as a confounder or an effect modifier in study 1 and 4.

These kinds of confounders include APOE status, blood pressure, cholesterol levels, lifestyle habits such as smoking, alcohol, physical activity, details on socioeconomic status, social activity, marital status, number of deliveries, and diabetes mellitus etc, which affect the overall health status of a person and either promoting or impairing cognitive and brain reserve thus acting as a barrier or a shortcut to the onset of debilitating illness like AD. We were able to account for certain comorbidities in two studies (1 and 4) which might take into account these potential confounders to some extent, such as the presence of cardiovascular diseases and cancer. In addition, the findings from study 2, where we had detailed information on lifestyle factors and socioeconomic position and education, were comparable to those from study 4. In study 3 that assessed the association between HT and cognitive decline, we were also able to account for APOE; this genetic factor did not have any effect on the results of that study. In summary, similar results were obtained in all of the studies making up this thesis, i.e longer HT exposure was associated with better cognitive outcomes. We interpret this to mean that the association of longer duration of HT and lower risk of AD was not due chance alone, but it cannot rule out the possibility that it could be explained by bias or confounding. Although we were able to adjust for multiple

lifestyle- and socioeconomic position- related confounders in studies 2 and 3, it could be that these confounders do not adequately capture the confounder that they hope to measure, or that there is some unmeasured confounder.

7 CONCLUSIONS

Based on the findings from the four studies included in this thesis, the following conclusions can be drawn

1. Postmenopausal gynecological surgery i.e. oophorectomy, hysterectomy, and hysterectomy with bilateral oophorectomy, was not a significant predictor of AD, irrespective of indication of surgery or HT use.

2. Postmenopausal HT was not significantly associated with AD and dementia in a longitudinal prospective cohort study while controlling for various midlife lifestyle and socioeconomic factors.

3. Long term postmenopausal HT (10-11 years) was associated with a lower risk of AD independent of oophorectomy and hysterectomy status.

4. Postmenopausal HT was not significantly related to cognitive decline although a protective association was detected between long term HT use and global cognition and episodic memory.

5. Short term postmenopausal HT use among women aged >65 years was associated with an increased risk of AD.