The studies, which were published between January 2005 and December 2015, included older adults aged 50 years and over. The inclusion criteria were: published in the English language in peer-reviewed journals, multimorbidity, house-bound elderly individuals, healthcare, home healthcare. The exclusion criteria were: review studies, qualitative studies, theoretical studies, studies of younger persons, studies published before 2005 and studies without multimorbidity, health, healthcare, health utilization or healthcare services in the title.

Electronic searches were performed in Academic Search Premier (290), Ovid Medline (1), PubMed (812), CINAHL (90) and ProQuest (772) for the period January 2005- December 2015 (see Table 1). A total of 1965 abstracts were read and 30 studies retrieved for further investigation. A manual search yielded an additional two studies. The retrieval and selection process, which is presented in Figure 1, resulted in a total of nine quantitative studies.

A systematic evaluation was conducted on the studies that met the inclusion criteria, which comprised an assessment of the validity and a systematic presentation of the findings. Similar evaluations have been done of in earlier systematic reviews [20] [21] . In this study the PRISMA checklist was employed [22] , Table 2.

The studies included a broad spectrum of content, measurements and statistical analyses across varying time periods, making a meta-analysis impossible [23] . An important aspect of systematic reviews is examination of the methodologies of the primary studies [19] [21] . The PRISMA checklist was used to review methodological aspects of the design, sample size, response rate, causality, measurements, generalization and ethical approval [20] , in addition to other methodological literature [19] [23] - [25] .

Design; Five of the studies were described as having a cross-sectional design [26] [28] - [30] . One study took the form of a survey [31] , while one was part of a longitudinal

ADL, Activities of Daily Living. B.IPQ, Brief Illnes Perceptions Questionnaire. BSSS, Assessment of subjective social support. CAPI, CATI, Computer Assisted Interviews. Nationwide population sample of men and women aged 65 years and older who participated in previous German Health Telephone Surveys. DSST, Digit Symbol Substitution Test. FLQM, Fragebogen zur Lebensqualität multimorbider älterer Menschen. HSRQOL, Health Related Quality of Life. ICPC, International Classification of Primary Care. HSU, Health Services Utilization. HCC, Health Care Costs. IPANAS, International Positive and Negative Affect Schedule. LINH, the Netherlands Information Network of General Practice. NDSS, National Diabetes Services Scheme. NSA, National Seniors Australia. PCG, Pharmacy-based Cost Group. OMAHA, population-based longitudinal epidemiological study of multimorbidity in population aged 65 years and older. PACIC, Patient Assessment of Chronic Illness Care. PCI, Proactive Coping Inventory. PHQ, Patient Health Questionnaire. SRH, Self-Rated Health.

design [30] . There was no information about the design in two of the studies [33] [34] . Causality; It was stated in one cross-sectional study that the design made it impossible to draw any causal conclusions [29] . Four of the other cross-sectional studies did not discuss any problems related to causality or internal validity [26] [28] [30] . Holzhausen et al. [32] reported that their longitudinal study provided opportunities for analyses of health trajectories, longitudinal relationships of outcome determinants and most importantly, causal relationships between conditions and trajectories over an 18 month period. Perrucio et al. [31] stated that their survey intended to convey theorized causal pathways and not longitudinal evolution. Selection bias; It was reported in one study that the findings could be subject to selection bias [29] . Nunes et al. [30] stated that the probability of selection bias in their population-based study was low due to the small number of drop outs and refusals [30] . Holzhausen et al. [32] recommended that a strategy of oversampling underrepresented, yet notably prevalent and politically important groups, could serve as a means of minimizing selection bias in future studies [32] . In one study it was suggested that the low response rate might have introduced selection bias because responders could have been more health conscious and hence more likely to report better self-rated health [31] . Three studies provided no information about selection bias [26] [28] [34] . However, Tooth et al. [26] stated that their sample was randomly selected with purposive oversampling in rural and remote areas [26] . The sample size of the studies ranged from N = 9958 to N = 229,493 (Table 2). In two studies the sample was described as relatively small [27] [33] . Four studies presented no information about selection bias related to sample size [26] [28] [31] [34] . Measurements; In one study it was stated that no standardized measure of multimorbidity exists for epidemiological surveys [29] . Different countries and research groups employ a variety of coding systems for chronic diseases, which means that results cannot be directly compared between countries [29] . One study reported the lack of a validated instrument for administration and that the assessment of subjective social support was restricted to one cohort [32] . Holzhausen et al. [32] stated that after completion of the analyses related to the validity and reliability of adaptations of existing measurement tools and a newly developed quality of life assessment, they would be able to provide a comprehensive set of instruments for face-to-face and telephone assessment in population-based health surveys. Seven studies provided no description of the validity or reliability of the measurements used [26] - [29] [31] [33] [34] . One study reported a limitation due to self-reported data [26] . Statistical analysis; Seven studies contained information about the statistical analysis used (Table 2), while two did not [30] [32] . Holzhausen et al. [32] explained that no power calculation was considered and that a statistician supervised, validated and, where necessary, corrected the data throughout the data collection process. Nunes et al. [30] described using Stata version 12. Generalization; In one study it was stated that generalizability was limited to elderly community dwelling individuals [31] . Another study used definitions to enhance the precision and generalizability of findings [33] . Three studies had samples that were predominantly female [26] [30] [31] . Seven studies did not mention anything about generalization to other older persons [26] - [30] [32] [34] .

Demographic data; All the included studies contained some demographic data (see Table 3).

The authors of the present review investigated the healthcare received by older patients with multimorbidity using a qualitative analysis to produce a thematic synthesis. The first author identified, grouped and summarized the findings as described by Pope et al. [35] . The thematic synthesis emerged by reading and re-reading the studies in order to synthesize the findings [19] [36] . Different patterns and concepts were identified across the data, leading to common meanings and concepts that were considered descriptions of the healthcare received by older patients with multimorbidity. The three authors discussed the emerging themes on several occasions before reaching consensus on the labelling.

Six studies stated that multimorbidity was high and constituted a significant problem in older adults [28] - [32] [34] . Orueta et al. [28] added that the distribution of the population by age group, as well as the average number of chronic diseases per patient, was 0.97 overall. Multimorbidity affected more than half of the population over 65 years and more than three quarters of those aged between 80 and 84 years [28] .

Cancer Surveillance Program (CSP); Breast Cancer Treatment Fund (BCTF); Self Reported Health (SRH).

Four studies revealed that multimorbidity was higher in women than in men [28] - [31] . Perruccio et al. [31] found that increasing age and female sex were associated with worse self-rated health. In contrast, Glynn et al. [33] revealed that gender had no significant effect.

Nunes et al. [30] stated that the higher frequency among women may be attributed to survival bias as men tend to die earlier and those who survive are usually the healthiest.

Three studies reported that patients with multimorbidity have high healthcare expenditure and costs [28] [29] [33] .

Orueta et al. [28] stated that multimorbidity was responsible for 80.57% of total healthcare expenditure. 23.61% of patients with multimorbidity accounted for 63.55% of total healthcare expenditure. The total healthcare cost per person was higher in more socially deprived areas. The same trend was seen in patients with multimorbidity. These differences were statistically significant in all health expenditure categories, both for the general population and for patients with multimorbidity. For patients with 10 or more chronic diseases, the only statistically significant difference was expenditure on medications [28] . Orueta et al. [28] found it was not only the number of chronic diseases that matter when investigating the increased cost, but which chronic diseases the patients suffered from and especially what was considered to be their primary disease. Chronic obstructive pulmonary disease, diabetes mellitus, heart disease and heart failure increased the cost in line with the number of other coexisting diseases. Orueta et al. [28] demonstrated that on average, the use of healthcare resources by patients with multimorbidity grew with the number of chronic diseases. This increase was not linear but tended to be progressive, which was the most common pattern in such patients. Exceptions were depression and anxiety, as patients with either of these conditions and with more than two other chronic diseases used less resource than if they only had the other chronic health problems [28] .

Bähler et al. [29] stated that outpatient services accounted for 50.8% of the total costs, inpatient services for 24.1% and medication for 25.1%. The mean total healthcare costs were 5.5 times higher in patients with multimorbidity compared to those with no or only one chronic condition. There were considerably higher mean costs for patients with multimorbidity in both the inpatient and outpatient subgroups. The total healthcare cost per year was related to nursing dependency and total health care cost in the preceding year, followed by the number of chronic conditions. Costs increased by 32.6% for each additional chronic condition and almost doubled in patients who required home care nursing services [29] . In the analysis by Bähler et al. [29] , the mean total costs were 5.5 times higher in the multimorbid compared to the non-multimorbid sample. The high standard deviations are consistent with the standard deviations in the cited articles, suggesting great variance and heterogeneity in the sample. Gender differences were found in health expenditure and costs, where costs were consistently higher in men aged 70 years or older, whereas in women costs decreased with increasing age [29] . A comparison between age and gender revealed that total costs increased by 9.10% in men aged 70 to 84 years compared to men aged 65 - 69 years [29] . A slight increase of 5.7% was found for the oldest age group. Overall, female gender was associated with lower costs, whereby costs decreased with increasing age for women aged 75 years and older [29] . The lowest total health care costs were therefore found in women aged 85 years and older, which represented a decrease of 12.1% compared to women in the youngest age group [29] .

Glynn et al. [33] added that primary and secondary care was influenced by the need for multiple healthcare professionals. The impact of gatekeeping on healthcare utilization merits further consideration. According to Glynn et al. [33] , there is a need to focus on coordinating the patient’s journey through the healthcare system, while at the same time promoting “self-management” among patients with multimorbidity. Pairwise comparisons showed that the mean total cost increased significantly for patients with a number of chronic conditions compared to those with no chronic conditions [33] .

Two studies focused on medication management problems in patients with multimorbidity [27] [34] . Jowsey et al. [27] revealed that an alternative view of healthcare complexity was medication management. The patterns in the targeted samples were broadly in the expected direction, with the unexpected values pertaining to those in the small sample categories. There was a strong relationship between medications and reported time use [27] . The one really clear statistically significant item was the number of conditions related to time use in all sub-samples. Illness management was time consuming for patients with multimorbidity [27] . Van Ooström et al. [34] demonstrated that patients with multimorbidity had a higher number of prescriptions and more referrals to specialized care than those with one or no chronic disease.

Two studies highlighted the social inequities in the health and care services [29] [30] . Bähler et al. [29] found a small but significant socio-economic gradient: the lower the purchasing power of a community, the lower the number of consultations. The decrease was 7.8% in patients living in a community with the lowest compared to the highest purchasing power. The lower the purchasing power of a community, the lower the costs per patient: the decrease was 3.4%, 6.6%, 9.5% and 13.4% with each reduction in the purchasing power quintile, compared to patients living in a community with the highest purchasing power [29] . Compared with the German-speaking part, the Italian-speaking part of Switzerland was associated with slightly lower costs of 2.3%. No significant differences in the French and the Romansh speaking parts were found [29] . Choosing a higher deductible was significantly and negatively associated with total costs, amounting to a decrease of almost one fifth, where no association was found between the total costs and being in a managed care model [29] . Nunes et al. [30] confirmed the impact of social inequities as the elderly participants in their study were poor, less well educated and had no private health plan. While acknowledging that these factors might have been confused with socio-economic indicators, Nunes et al. [30] were of the opinion that an adjusted analysis would not make sense in view of the aims of their study and the fact that their sample had more diseases and greater social and financial vulnerability. Thus, health interventions related to the treatment and monitoring of chronic conditions should prioritize these individuals [30] . The occurrence of multimorbidity was higher among those with black or brown/yellow/indigenous skin, those with less education, less purchasing power, those who are bedridden, those who did not have an individual health plan, those who had medical consultations and visited emergency services, those who had been hospitalized and those living in community care [30] areas.

Four studies revealed a trend of complex healthcare and/or consultation needs in patients with multimorbidity [29] [30] [33] [34] . Bähler et al. [29] reported that complex home nursing services and being in a managed care model were significantly associated with a higher number of consultations, an increase of 11.5% and 2.9% respectively. Living in the French or Italian speaking parts of Switzerland was associated with a lower number of consultations [29] . The effect of age on the increase in the number of consultations was greater in females compared to males. However, the number of consultations increased with advancing age in men, but the increase was less in women, although the results did not differ significantly for the last age group (women aged 84 to 89 years compared to 65 - 69 years). Women aged 85 years or older exhibited a significant decrease of 5.1% in the number of consultations, whereas an increase of 7.6% was found in men of the same age. 98.5% of patients with multimorbidity had at least one consultation in 2013 compared with 68.7% of patients without multimorbidity [29] . The sample of patients with multimorbidity was almost twice as likely to have had a consultation with a primary care physician and with a specialist. Approximately 22% of the patients with multimorbidity were seen by more than one primary care physician and over 54% consulted more than one specialist over the same time period. Patients with multimorbidity were 5.6 times more likely to be hospitalized compared to the patients without several diseases [29] . The mean number of consultations per year amounted to 15.7 in the patients with multimorbidity compared to 4.4 in the patients without multimorbidity [29] .

Nunes et al. [30] found that the number of individuals requesting healthcare was significantly higher in the multimorbid group compared to the non-multimorbid sample. The most prevalent dyads of morbidities were high blood pressure and spinal column diseases (23.6%) and high blood pressure and heart problems (22.3%) [30] . Four of the dyads did not have a frequency that was statistically higher than expected. In the triads, this only occurred with the HBP/spinal column disease/cognitive impairment triplet. In the dyads, the highest ratio between observed and expected frequency was found for rheumatism and spinal column disease [30] .

More specifically, Glynn et al. [33] found a significant interaction between age and gender, where the effect of age on the increase in the mean number of primary care consultations was larger in females compared to males. There is evidence of a linear trend for primary care consultations and hospital admissions, while a trend also exists for hospital outpatient visits [33] .

Van Ooström et al. [34] found that patients with multimorbidiy had significantly more face-to-face consultations, telephone consultations, home visits, diagnostics or minor surgery and contacts for prescribing medications in general practice than patients with one or no chronic disease. The number of chronic diseases was linearly associated with the number of contacts in general practice. The number of general practice contacts increased with the number patients with multimorbidity [34] . Patients with heart failure and comorbid diseases had the most contacts with general practice. The number of contacts per disease decreased with each extra disease. van Ooström et al. [34] demonstrated a significantly lower number of contacts in general practice for those with specific combinations of two chronic diseases than would be expected on the basis of the contact frequency associated with each individual chronic disease. The confidence interval indicated that the number of contacts per year was lower than expected on the basis of the number of yearly contacts for diabetes and for coronary heart disease separately. There were no disease pairs where the number of contacts was higher than expected on the basis of the contact frequency associated with the separate diseases [34] .

Tooth et al. [26] developed and validated multimorbidity indexes to predict mortality, visits to general practitioners and specialists, hospitalization and Ability for daily living (ADL) in a population-based sample of older Australian women. Seven types of chronic condition such as cancer and stroke were significantly associated with the risk of mortality, different use of healthcare services and predicted a high frequency of contact with general practitioners [26] . Seven of the 19 explanatory variables were statistically significant in the stepwise proportional hazards regression analysis of mortality; these were heart disease, stroke, low iron, bronchitis/emphysema, diabetes, all forms of cancer other than skin and Alzheimer’s disease [26] . Several chronic conditions were strongly associated with all health care service use outcomes. The presence of Alzheimer’s disease was very strongly associated with the need for assistance. The regression coefficients from these models were used to produce the weights for each of the outcomes, after which the weights were employed to construct weighted scores [26] . Mortality had an essentially linear relationship with both the weighted and unweighted scores [26] .

Risk of bias will be discussed in this section, as suggested by Schneider et al. [24] and Polit and Beck [19] . Bias is defined as influence that produces a distortion of the results, thus threatening validity and trustworthiness [19] .

The design of five studies was described as cross-sectional, one study took the form of a survey, while one study was part of a longitudinal design (Table 2). Cross-sectional studies, surveys and longitudinal studies can employ an observational design. Two studies contained scare information about the design [32] [34] . Observational designs can have limitations in their ability to determine “causality” ( [24] , p. 163), which was mentioned in one study [29] . The implications of studies with an observational design can be weakened by the fact that the participants may change over time, which Polit and Beck ( [52] , pp. 246-247) called the “history and/or maturity threat”. A longitudinal design typically requires the participants to retrospectively summarize the extent to which they employed various strategies over a long period and is more likely to produce inaccurate results due to memory bias, compared to measurements that focus on a shorter timeframe. However, none of the included studies stated that the retrospective design could decrease validity and/or trustworthiness. A prospective design starts with a presumed cause and continues with a presumed effect [52] . It was stated in two studies that self-reported data were used [26] [30] . Tooth et al. [26] mentioned that the limitations of self-reported data must be acknowledged. Self-reports can increase the likelihood of response bias [25] . The two studies that employed self-reports described this method as a limitation that can decrease validity and lead to bias [26] [30] . There was no reference to methodology in any of the included studies.

Statistical power depends on sample and effect size, while non-significant implications must be interpreted with caution [24] [25] . Failure to show significance was described as due to lack of power. However, this information was not related to methodological literature. Two studies described having a relatively small sample size (Table 2), but did not mention selection bias as a limitation. The use of only one group was not referred to as decreasing validity in these two studies. In the opinion of the authors of this review (ALH, AB, ES), the lack of common measurements and definitions of the concept of multimorbidity is the most serious limitation.

A limitation of this review is the fact that one of the studies comes from Switzerland, one from Ireland, one from Germany, one from The Netherlands (four from Europe), one from USA, one from Canada and one from Brazil (three from south and north America), and two from Australia (see Table 2). The different cultures in these parts of the world must be taken into consideration; for example, the various countries seem to have different way of registering and defining multimorbidity in clinical settings. Further studies are required, as is the development of measures and common strategies for deciding which chronic diseases and conditions should be included in samples. Researchers need to strengthen the design, validity and reliability of quantitative studies on multimorbidity.