Summary of findings
• Prevalence. BPS occur in non-demented, cognitively impaired and demented populations, but there are large variations in reported prevalence estimates.
• Biological associations. Biological factors associated with BPS include health status, homocysteine levels and vascular disease.
• Risk factors. Associations have been found with psychosocial environment and dementia severity. Results are inconsistent, especially for socio-demographic factors, and differ between symptoms.
• Outcomes and care. BPS have been associated with increased burden of care, decreased caregiver's general health and increased caregiver depression scores, but these conclusions come from a limited number of studies. Findings in relation to BPS and institutionalisation, quality of life and disease outcome are generally inconsistent.
Limitations of reviews
Over 97,000 articles have been published on BPS in recent years (up to July 2011). However, in total, only 36 systematic reviews were identified, published from 1989 to 2011, covering 6 to 244 papers, with only 10 reviews, including a meta-analysis. These covered a wide area of research and included studies with large differences in population characteristics, recruitment and definition of BPS. Depressive symptoms were most widely reviewed. Other symptoms, such as apathy, irritability, wandering and elation, are typically ignored, especially in studies of the older population. Aggression, psychosis and wandering have been identified as the BPS that are most difficult to cope with by caregivers [61
]. Not all areas of BPS research have yet been adequately reviewed. For example, no review has yet appeared on the neuropathology underpinning BPS, although there are several recent studies on this subject and many possible biological mechanisms may underlie BPS, such as Alzheimer's disease pathology, lower neuronal counts, neurotransmitter changes, genetic risk factors and abnormal neuroendocrinology [62
]. A high quality systematic review of the biological underpinnings of BPS would give direction to further research, particularly with regard to treatment development.
The quality of the included reviews as measured with the AMSTAR tool was generally low. All except one did not report if the research question and inclusion criteria were established before the conduct of the review, or they did not provide a research question and/or inclusion criteria. Many reviews did not have two independent researchers who selected the studies and extracted the data, or did not report this. None of the reviews reported the potential conflict of interest of the included studies. Furthermore, the inclusion of grey literature and if the search was supplemented by consulting current contents, reviews, textbooks, specialised registers or experts were rarely reported.
Better and more systematic reporting would make it easier to compare the characteristics, results and strengths and limitations of reviews, and the use of reporting checklists, such as PRISMA, is recommended [65
]. More details on the populations covered by the review and the characteristics of included studies would be particularly helpful.
Recommendations for future research made by the reviews were variable, with many recommendations made only by one or two reviews. Only more general recommendations were made by several reviews. This has been previously reported by Clark et al.
, who examined 2,535 Cochrane reviews and found that the characterisation of the needs for future research was less than explicit [66
]. In addition, it has been previously found that reviews tend to be too optimistic when drawing conclusions from their results [67
]. It has been recommended that research gaps should be identified more systematically, rating the reasons of research gaps in terms of population, intervention, comparison, outcome and setting (PICOS), including insufficient information, biased information, inconsistency or not the right information [69
], although this tool has been designed for reviews of intervention studies and may not be suitable for reviews of observational studies.
Limitations of original studies
The reviews have included individual research studies with a large degree of heterogeneity in study design (for example, definition and measurement of BSP, level of cognitive impairment, population characteristics and recruitment of participants), making cross study comparisons difficult. Many different instruments are used to measure BPS, including the Neuropsychiatric Inventory (NPI) [70
], CERAD-BRSD [71
] and the Behavioral Pathology in Alzheimer's Disease scale (BEHAVE-AD) [72
]. Symptom-specific instruments are also used, including the Geriatric Depression Scale (GDS) for depressive symptoms [73
]. Some instruments use self-ratings (for example, GDS), while others are based on a caregiver interview (for example, NPI). Across instruments there is no consistency in the definition and severity of the symptoms.
Some studies focused on the general older population, while others selected only those with dementia or MCI. Even within impaired groups large differences exist in the measurement of cognitive function. Some study populations were divided by cognitive status (for example, normal, mild or other cognitive impairment and dementia) using diagnostic criteria applied independently of BPS. It may be that the phenotypes of BPS (for example, the causes of depression or agitation) have a different basis at different stages of cognitive impairment and, therefore, studies using different groups will have discrepant findings. Other features, such as the age and gender of participants, and the setting where they were recruited, also vary. If samples are not representative of the population (either that of older people in the community or people with dementia) conclusions may be difficult to generalise.
Limitations of our review of review
There is no Mesh or Emtree search term for BPS, so search terms for individual symptoms and text searches had to be combined. As many different terms are used to describe BPS, it is possible that we may have missed relevant reviews, though we have checked all reference lists of relevant papers. One author undertook data selection and extraction.
Many different definitions of BPS have been used in the literature. Here we used symptom definitions as used by the most commonly used instruments to measure BPS, including the NPI. However, symptom definitions may overlap and there was heterogeneity in symptom definitions that were used by the reviews. Depression is especially problematic. All of the reviews included here that studied depression included studies investigating major depression as well as studies investigating depressive symptoms. In the original studies, decisions about depression can be made at recruitment, when applying the inclusion and exclusion criteria or during measurement at baseline and follow-up. In some studies, only people with or without depressive symptoms or major depression are recruited, or participants with major depression can be excluded from the study at the participant selection stage. At the data collection stage, the definition of depression can include both major depression and minor depression using a cut-off score (for example, a CES-D score higher than 16); it can include both depressive symptoms and major depression separately (for example, DSM definitions for major and minor depression or CES-D scores 16 to 20 and CES-D higher than 20), or either depressive symptoms only (for example, using the Neuropsychiatric Inventory to measure depressive symptoms in the context of BPSD) or major depression only (for example, DSM definition for major depression).
By reviewing reviews and not the original studies, it is likely that noise has been introduced. Furthermore, with the diverse and generally qualitative outcomes of the majority of reviews, we were unable to combine results quantitatively and, therefore, a meta-meta-analysis was not possible.
Recommendations for future research
Recommendations for future reviews on BPS:
• More reviews on BPS are needed, for example, on neuropathological associations
• Improve the quality of reporting of reviews
• Choose reviews that include all BPS not just symptoms of depression/psychosis
• Clearer reporting, using a checklist tool, for example, PRISMA
• Clearer and more specific recommendations for future research, for example, PICOS if applicable
Recommendations for original research on BPS:
• Prospective longitudinal studies
• Large sample size
• Standardised instruments for BPS
• Wide age range, including oldest old
• Improved comparability of results; report study characteristics clearly