Study inclusion criteria were: (i) sample size of 100 or more; (ii) participants aged 50 and older, (iii) data on populations both with and without moderate to severe cognitive impairment; (iv) a valid and reliable measure of dementia or moderate-to-severe cognitive impairment; (v) at least one other CC or geriatric syndrome or measure of comorbidity; and (vi) a description of the prevalence or effect of cognitive impairment on co-occurring CCs.

To define cognitive impairment, the article must include a valid and reliable clinical diagnosis of dementia (e.g., the Diagnostic Statistical Manual criteria, fourth edition (DSM-IV) (American Psychiatric Association, 2000); the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria for Alzheimer’s disease (McKhann et al., 1984); or the National Institute of Neurological Disorders and Stroke and Association Internationale pour la Recherché et l’Enseignement en Neurosciences (NINDS-AIREN) criteria for vascular dementia (Erkinjuntti, 1994) or valid and reliable measures of moderate-to-severe cognitive impairment that captured memory and at least one other cognitive domain such as executive function or language. These could include both multidimensional measures (e.g., Mini-Mental Status Examination (MMSE)) (Folstein et al., 1975) or multiple individual measures of cognitive function with at least one focused on memory. The review focused on moderate-to-severe cognitive impairment, including persons with and without official dementia diagnoses. We included the latter group given the high rate of persons without a formal diagnosis; currently, less than half of those in high income countries and less than 10% in low-income and middle-income countries have received a dementia diagnosis (Prince et al., 2013).

Using the definition of CCs as “conditions that last a year or more and require ongoing medical attention and/or limit activities of daily living (ADL’s)” (Warshaw, 2006; US DHHS, 2010) the expert panel created a list of common CCs (Table 2) from The NIH Cognitive and Emotional Health Project (Hendrie et al., 2006), the U.S. DHHS multiple chronic conditions (MCC) Strategic Framework (US DHHS, 2010), and disease categories in PubMed. Geriatric syndromes (e.g., frailty and polypharmacy) were included as well given their large impact on quality of life and disability (Inouye et al., 2007). Geriatric syndromes are conditions that are common in older adults but do not fit into discrete disease categories (Ahmed et al., 2007; Inouye et al., 2007; Chaudhry et al., 2010). To be included in this review, the article had to include people living with at least one CC condition or geriatric syndrome (Table 2), or evidence of co-morbid conditions via a valid and reliable measure of comorbidity (e.g., Charlson’s Comorbidity Index) (Charlson et al., 1987).

Articles included both community-based and institutionalized populations from around the globe. We excluded articles not published in English, that only used an ICD diagnosis for cognitive impairment or dementia, that defined cognitive impairment as delirium, traumatic brain injury, developmental disorders, or MCI, and that reported only reviews, commentaries, or meta-analyses.

We used a two-step screening process to assess whether articles met inclusion criteria. We first reviewed titles and abstracts. Articles that could not be included or excluded based on abstracts were retrieved to examine the full text in a second screening phase. Screening of titles and abstracts was conducted jointly by members of the research team (LS, LB, and CC) until 80% agreement was reached; subsequent title and abstract screening was conducted separately. Articles were screened by individual reviewers. A different research team member abstracted data from accepted articles to double check whether an article met eligibility criteria. The reviewers met weekly to discuss title/abstract and article screening. The principal investigator (MS) made final determinations when questions or disagreements between reviewers could not be resolved.

We used a standardized form to systematically collect data from each article, including study design, sample size, setting, co-occurring chronic disease, outcome measures, results, and indicators of study quality (e.g., adjustment for possible confounders). Data were compiled in summary tables for evidence rating (see Appendices). We grouped articles into chronic condition-outcome pairings to categorically rate the evidence, and panelists rated the overall evidence within each group; for example, rating the evidence for how cognitive impairment impacted mortality for persons with multiple CCs, stroke, or depression.

Expert panel members rated both the quality and strength of evidence across articles in each chronic condition-outcome pairing using a qualitative approach from previous systematic literature reviews (Frederick et al., 2007; Snowden et al., 2011). Indicators of quality included sample size and generalizability, appropriate measurement of cognitive impairment and of CCs, use of multivariable analysis, and controlling specifically for age, gender, and education. For quality rating, panel members independently rated each condition-outcome pairing as “good”, “fair”, or “limited”.

Indicators of strength of evidence included the study quality and design, the number of studies in the chronic condition-outcome pairing, consistency across studies, and statistical findings. For strength of evidence ratings, the panel members independently rated each condition-outcome pairing as “strong”, “sufficient”, or “insufficient” (recording whether the insufficient rating was due to an insufficient number of at least fair quality studies or a sufficient number of available studies but inconclusive data).

Final determination of quality and strength of evidence was based on 80% agreement among panel members (i.e., agreement among six out of the seven panelists). The panel met to discuss areas of disagreement, and panel members were allowed to change their votes after the discussion; however, to reduce persuasion bias, they were not required to reach consensus.

The panel also identified and prioritized the areas for which there was insufficient evidence. We used Q-sort ranking (Brown, 1996), a method for rank-ordering items, to prioritize the gaps in which panelists assigned a rank (“5”=highest priority to “1”=least priority) to each of the 42 outcomes and CCs/geriatric syndromes with insufficient evidence in order of priority for further research (“5”=highest priority to “1”=least priority). Using the Q-sort method, informants were given a limited, normally distributed allotment of 1s, 2s, 3s, 4s, and 5s to create a normal distribution of rankings.

Twenty-nine studies (Kukull et al., 1994; Arfken et al., 1995; Bruce et al., 1995; Gale et al., 1996; Agüero-Torres et al., 1999; Foley et al., 1999; Kammoun et al., 2000; Helmer et al., 2001; Stump et al., 2001; Freels et al., 2002; Ganguli et al., 2002; Feil et al., 2003; Nguyen et al., 2003; Tschanz et al., 2004; Cacciatore et al., 2005; Fitzpatrick et al., 2005; Magaziner et al., 2005; Bursi et al., 2006; Guhne et al., 2006; Llinàs-Regla et al., 2007; Lyketsos et al., 2007; Meerman et al., 2008; Rothman et al., 2008; Zekry et al., 2009; Lavretsky et al., 2010; Wang et al., 2010a; Gombojav et al., 2011; Millán-Calenti et al., 2011; Nikolova et al., 2011) compared mortality rates for persons with multiple CCs and with and without moderate-to-severe cognitive impairment or dementia. Seventeen used community-based samples and 12 used clinic-based samples. Most studies had a sample size greater than 500 (79%) with 52,644 subjects across all studies. A majority of studies used a multivariable analysis (90%) and almost half of these (46%) adjusted for age, gender, and education (the remaining studies adjusted for just one or two of these demographic covariates).

Mortality was measured in terms of mortality rates, risk for mortality (e.g., hazard ratios (HRs)), or survival rates. Percent mortality ranged from 40.2% to 82% for persons with moderate-to-severe cognitive impairment and 16.0% to 37.7% for persons without significant cognitive impairment (Stump et al., 2001; Nguyen et al., 2003; Tschanz et al., 2004; Guhne et al., 2006; Llinàs-Regla et al., 2007; Lavretsky et al., 2010; Wang et al., 2010a). Six studies (Bruce et al., 1995; Stump et al., 2001; Fitzpatrick et al., 2005; Llinàs-Regla et al., 2007; Meerman et al., 2008; Lavretsky et al., 2010) reported unadjusted relative risks (RR) for mortality (95% confidence interval (CI)), ranging from 2.3 (1.7–3.2) (Llinàs-Regla et al., 2007) to 7.4 (4.9–11.4) (Lavretsky et al., 2010).

Fourteen studies (Helmer et al., 2001; Stump et al., 2001; Freels et al., 2002; Feil et al., 2003; Nguyen et al., 2003; Tschanz et al., 2004; Fitzpatrick et al., 2005; Bursi et al., 2006; Guhne et al., 2006; Llinàs-Regla et al., 2007; Meerman et al., 2008; Rothman et al., 2008; Lavretsky et al., 2010; Gombojav et al., 2011) reported the adjusted HR for significant cognitive impairment on risk of mortality (95% CI) ranging from 1.5 (1.1–2.1) (Rothman et al., 2008) to 2.99 (2.53–3.53) (Tschanz et al., 2004); covariates may have included demographics, CCs, and six other frailty criteria (Rothman et al., 2008). The percent attributable risk (PAR%) of death related to dementia diagnosis ranged from 11.8% (Guhne et al., 2006; Llinàs-Regla et al., 2007) to 16.6% (Tschanz et al., 2004). Tschanz (Tschanz et al., 2004) and Feil (Feil et al., 2003) compared mortality RR and PAR for dementia versus other CCs: In addition, two studies (Feil et al., 2003; Tschanz et al., 2004) found that dementia or moderate-to-severe cognitive impairment had a greater RR of mortality than other CCs (e.g., an RR two to three times higher) and higher mortality PAR% than other CCs.

In addition, persons with moderate-to-severe cognitive impairment had shorter average survival times, ranging from 3.0 to 7.1years compared with 4.0 to 11.0years for persons without cognitive impairment (Agüero-Torres et al., 1999; Fitzpatrick et al., 2005; Guhne et al., 2006; Lavretsky et al., 2010; Wang et al., 2010a). Persons with moderate-to-severe cognitive impairment also had lower survival rates: the adjusted 5-year survival for persons with moderate-to-severe cognitive impairment was 16.1% versus 28.5% for normal cognition (Wang et al., 2010a); unadjusted survival rates were 22% versus 39% (Bursi et al., 2006) and lower (poorer) scores on the MMSE decreased likelihood of survival (Bruce et al., 1995).

Several studies reported non-significant or negative relationships. For example, in 2005, Magaziner et al., (2005) reported an unadjusted RR for mortality of 0.61 (0.53–0.71) and adjusted RR for mortality of 0.63 (0.51–0.77) for persons with dementia and multiple CCs. This study included persons living in nursing homes (N = 2,153, mean age > 80), almost half of which were diagnosed with dementia using DSM criteria. In this study, the group without dementia included people who were difficult to diagnose.

The expert panel also found sufficient evidence that moderate-to-severe cognitive impairment increases mortality in persons with Parkinson’s disease (Ebmeier et al., 1990; Marder et al., 1991; Mitchell and Rockwood, 2000; Levy et al., 2002; Parashos et al., 2002; Buter et al., 2008; Lo et al., 2009; Hobson et al., 2010, N = 4,513) and stroke (Tatemichi et al., 1994; Desmond et al., 1998; Desmond et al., 2002; Liebetrau et al., 2003; Melkas et al., 2009; Oksala et al., 2009, N = 2,302). These were the only specific co-occurring CCs for which sufficient evidence was found to examine mortality outcomes. Other articles included people with dementia or moderate-to-severe cognitive impairment from various etiologies (stroke, Parkinson’s disease, and other).

Three studies (Smith et al., 2000; Magaziner et al., 2005; Rothman et al., 2008, (N = 3,423) examined length of stay in an institution for persons with multiple CCs (Inouye et al., 2007). In 2008, Rothman et al., (2008) reported an adjusted HR for long-term nursing home stay of 3.7 (95% CI) 2.5–5.4) for persons with cognitive impairment (MMSE < 24) versus those without impairment. In 2000, Smith et al., (2000) found an unadjusted median length of stay in nursing homes for those with dementia (diagnosed using DSM-III-R criteria) of 946 days compared with those without dementia at 579days. Lastly, Magaziner and colleagues reported a decreased likelihood of being discharged to home (adjusted RR of 0.23 (95% CI 0.17–0.31)) for those with dementia (DSM-III-R criteria) compared with those without.

Nineteen studies (Agüero-Torres et al., 1998; Zhu et al., 1998; Agüero-Torres et al., 2002; Covinsky et al., 2003; Lyketsos et al., 2005; Magaziner et al., 2005; Lyketsos et al., 2007; Cankurtaran et al., 2008; Rothman et al., 2008; Zekry et al., 2008; Huang et al., 2009; Sousa et al., 2009; Stewart et al., 2009; Zekry et al., 2009; Feng et al., 2010; Wang et al., 2010b; Gombojav et al., 2011; Millán-Calenti et al., 2011; Nikolova et al., 2011) (N = 37,466) examined the impact of significant cognitive impairment and co-occurring multiple comorbidities on function. Function was typically defined as basic activities of daily living (ADL’s) or instrumental ADL’s (IADLs). Nine studies included samples with a dementia diagnosis and seven included institutionalized persons. Half of the studies used multivariable analysis, with half of these adjusting for age, sex, and education. Persons with dementia and multiple CCs had significantly more IADL and ADL impairments than those without significant cognitive impairment.

Seven studies (Freels et al., 2002; Lopponen et al., 2005; Bursi et al., 2006; Rastas et al., 2007; Andersson et al., 2008; Barzilay et al., 2008; Cankurtaran et al., 2008) (N = 5,245) examined the degree to which dementia affected cardiovascular medication use (other than that for hypertension) in persons with co-occurring cardiovascular disease (CVD). Only one (Barzilay et al., 2008) found statistically significant differences between people with dementia (VaD, AD, and not otherwise specified) and use of CVD medications such as beta-blockers, ACE inhibitors, or aspirin.

In addition, the expert panel agreed on sufficient evidence that moderate-to-severe cognitive impairment had no impact on hypertension medication use in people with co-occurring hypertension (Freels et al., 2002; Hanon et al., 2006; Barzilay et al., 2008; Vinyoles et al., 2008; Huang et al., 2009; Stewart et al., 2009; Gombojav et al., 2011) with five of the seven studies (Freels et al., 2002; Hanon et al., 2006; Huang et al., 2009; Stewart et al., 2009; Gombojav et al., 2011) (N = 8,236) reporting no statistically significant differences for persons with and without moderate-to-severe cognitive impairment.

The expert panel identified 55 pairings with insufficient or no evidence (Table 4). These included 18 outcomes (e.g., cost and function) and 10 CCs or geriatric syndromes (e.g., cancer and chronic obstructive pulmonary disease (COPD)) with insufficient evidence, and no evidence was found for the impact of significant cognitive impairment on 14 additional conditions/syndromes: asthma, arthritis, bone diseases, chronic kidney disease, chronic pain, dental problems, other chronic mental illness, neurological conditions, sleep disorders, substance abuse, risk factors if another CC was present (e.g., obesity and smoking), frailty, functional impairments, and urinary incontinence.

After the expert panel prioritized these gaps, the following were identified as highest priority for future research: service utilization-hospitalizations and disease-specific outcomes, and diabetes, chronic pain, CVD, depression, falls or fractures, stroke, and multiple CCs (the presence of moderate-to-severe cognitive impairment and more than two other CCs).

We were recently asked to update the literature review to include more recent articles published between mid-2011 and June of 2016. Twenty-two additional articles met our criteria and two new chronic condition-outcome pairings had sufficient evidence. The first looked at the impact of cognitive impairment and co-occurring CVD on mortality. Two studies (O’Donnell et al., 2012 and Huijts et al., 2013) found an over 50% increase in mortality risk for people living with moderate-to-severe cognitive impairment (adjusted HRs 1.53 (95% CI 1.02–2.30) to 1.68 (1.49–1.90)). The second new outcome was disease-specific; namely that living with dementia and diabetes increased the risk for severe hypoglycemia compared with persons living with diabetes and without dementia (Abbatecola et al., 2015; Prinz et al., 2015). Fourteen new articles (Chang et al., 2012; Chen et al., 2014; Huijts et al., 2013; Llibre et al., 2014; Katsoulis et al., 2014; Lee et al., 2015; Mignardot et al., 2014; Murao et al., 2014; O’Donnell et al., 2012; Park et al., 2016; Sanyal et al., 2014; Warchol-Celinska et al., 2015; van Asch et al., 2013) fell into existing chronic condition-outcome pairings and did not change the quality or strength of the evidence, and four articles (Hawkins et al., 2012; Jacobs et al., 2012; Hajduk et al., 2013; Provencher et al., 2015) were found in separate areas where there was previously no evidence.