This study revealed five specific clinically consistent patterns of multimorbidity in the adult population: cardio-metabolic, psychiatric-substance abuse, mechanical-obesity-thyroidal, psychogeriatric and depressive. Two of them were found to evolve throughout life and differed in their presentation for men and women (i.e., cardio-metabolic and mechanical-obesity-thyroidal), three of them affected both sexes (i.e., cardio-metabolic, mechanical-obesity-thyroidal and psychogeriatric), and two of them were present exclusively in men or women (i.e., psychiatric-substance abuse and depressive, respectively) ().
Identified multimorbidity patterns and their prevalence within each age and sex group.
Those two patterns suggesting an evolution throughout life were composed mainly of risk factors in the youngest age group, of organ disorders in the middle-aged group, and of various disease-related complications in the eldest group.
Although there are many examples in the literature of the association between the specific diseases that comprise these patterns, there have been very few population-wide published studies on multimorbidity including the non-elderly population.
This multimorbidity pattern was present in both sexes and for all of the age ranges analysed. Although it was especially prevalent among individuals 65 years of age or older (i.e., one in three women and one out of every five men in this age group), the high prevalence of this pattern in the middle-aged groups should be noted (i.e., 22% in women and about 10% in men).
Its development in men and women was consistent with the disease pathophysiology. In young patients, it was manifested as diabetes, hypertension, obesity and dyslipidaemia 
, and its progression over time was associated with the expected complications for this type of diseases, which were primarily cardiac in nature.
In men, this pattern presented at a young age and with a possible common pathophysiological basis of insulin resistance, obesity and their associated inflammatory processes, and negative lifestyles (e.g., physical inactivity, poor diet, etc.). During middle age, ischemic heart disease, atherosclerosis, myocardial infarction, arrhythmias, dyslipidaemia (with a factor score very close to the established cut-off point), substance abuse, COPD and chronic liver disease were also observed within this pattern. Although this study did not have access to specific information regarding tobacco smoking, it is likely that its prevalence was high in those cases where the abuse of other substances was also high (this information was available). The alcohol-tobacco association could have been the probable cause of the diseases observed, such as chronic liver disease and other related associations. In addition, there are possible underlying iatrogenic causes due to the use of drugs such as statins and fibrates, which can cause the elevation of transaminases at medium and high dosages 
. Dyslipidaemia was present in middle-aged individuals but was absent in men 65 years of age or older, which is inconsistent with the clinical picture observed for this age group. It is likely that this absence was due to an underreporting bias, given the greater relative severity of other diseases in this group. The presence of heart failure in the group of 65 years and over could be seen as the final link in the natural history of this cardio-metabolic disease. In our study, gout was added to this pattern, and its association with metabolic syndrome has recently been demonstrated 
The presence and the evolution of this cardio-metabolic pattern revealed differences between women and men. In women, although hypertension, obesity and dyslipidaemia were present at a young age, diabetes did not appear until the age range 45–64. Moreover, in this second age group, the expected cardiac complications were not observed, contrary to what happened in men of the same age. Oestrogenic protection and reduced smoking have been previously reported as the primary reasons for reduced cardiac complications in women 
. We propose a hypothesis for this study, which has also been examined by other authors 
, holding that metabolic syndrome should not be considered as a single, whole disease but rather that the obesity component should be considered the index factor, which when present at an early age of life, activates the immune system in both men and women and leads to the development of the so-called metabolic syndrome. Type 2 diabetes, therefore, would not be part of the “whole” (metabolic syndrome) but would rather be the consequence of the initial condition (obesity). This, together with the protective effect of oestrogen, would be the reason why Type 2 diabetes appeared in women after the age of 45 in this study. However, in the oldest age group studied (65 years of age and over), this pattern was presented similarly in men and women, amplifying this syndrome along with the various manifestations of heart disease, including ischemic heart disease, arrhythmias and congestive heart failure. Additionally, dyslipidaemia was absent in this group, as observed in men of this age, and this was likely due to the disease recording habits of professionals. In women, heart failure may have also served as the final link in the natural history of this cardio-metabolic pattern.
Psychiatric – Substance abuse
This multimorbidity pattern appeared only in young men and was found to affect 2% of the individuals studied (n
837). This pattern consisted of psychopathological processes, such as psychosis and neurosis, which are both likely related to the toxic substance abuse that is also present within this pattern and that commonly affects men at this stage of life 
. Psychoactive drugs may act as an agglutinative factor leading to psychosis, which in its early stages favours substance abuse 
. This clinical picture also involves obesity, which is a consequence of this poor lifestyle as well as the use of new-generation antipsychotic medications 
. This causal hypothesis is supported by the fact that this pattern did not appear among women, where toxic substance abuse occurs less frequently 
In men, this was a complex pattern that could also be denominated mechanical-obesity-thyroidal. It affected 5% of middle-aged individuals and was found to decrease to 2% among individuals aged 65 and over. For men aged 45–64, it is likely that obesity acted as an index disease that favoured the emergence of mechanical disorders due to excess body weight, such as arthropathy, cervical and low back pain, varicose veins of lower extremities, and gastro-oesophageal reflux 
. An increased risk for benign prostatic hypertrophy has also been found in obese men 
. However, this hypothesis should be further supported since, unexpectedly, obesity does not appear in men over 65. Instead, the factor score for thyroid problems became very close to the established cut-off point in this age group. The latter could explain some of the associations and interactions between the various diseases found within this pattern. It should finally be mentioned that for patients 65 years of age or older, this pattern shared clear similarities with fibromyalgia 
In women, the mechanical pattern was manifest somewhat differently than in men. It appeared early in life and it was never associated with obesity but rather with thyroid problems, likely in the form of hypothyroidism 
. While the prevalence of this pattern was 3% among women of the 15 to 44 age group, it was very high among women aged 45 to 64 affecting more than one in five women (22%). At this age, the pattern was amplified by neurogenic disorders, such as anxiety and dermatitis. Finally, for women in the 65 and over age group, the pattern was presented in its most amplified state (i.e. arthropathy, cervical and low back pain, varicose veins of lower extremities, gastro-oesophageal reflux, osteoporosis and neurogenic disorders) and was associated with various other disorders, such as dyslipidaemia or diverticulosis, for which the association is difficult to explain according to available knowledge.
This multimorbidity pattern appeared in individuals 65 years of age and older and was the second-most prevalent pattern after cardio-metabolic. It was more common in women (17%) than men (14%) and was manifest in the form of dementia, behavioural problems, Parkinson's disease, osteoporosis, chronic skin ulcers and iron deficiency. Heart failure and stroke showed factor scores very close to the established cut-off-point and are among the leading causes of dementia 
. Dementia can also be caused by Parkinson's disease, and the antipsychotic treatment of the behavioural disturbances in patients with dementia can cause Parkinsonism 
The presence of dementia and Parkinson's disease together with age-related osteoporosis can lead to falls, fractures, the immobility of the patients and the appearance of skin ulcers caused by bed rest among these patients 
In women, this pattern was slightly different than in men: dementia, cerebrovascular disease and chronic skin ulcers were present. In addition, heart failure, iron deficiencies and cardiac arrhythmia (mostly in the form of atrial fibrillation) were present in this pattern, and these were clearly associated with and were the cause of cerebrovascular disease 
. Parkinson's disease, osteoporosis and behavioural problems disappeared in women with respect to men.
This multimorbidity pattern consisted of only two conditions: depression and behavioural disorders (mainly insomnia). These were strongly associated with one another and were present only in women aged 45–64 and 65 years and over. This was the least prevalent of the five patterns described, as the frequency was less than 0.2% in both age groups. The association between these two clinical conditions has been widely described 
, and it has also been shown that untreated anxiety disorders can develop into depression 
. It is noteworthy that anxiety was present in our study population as part of the thyroid pattern in both men and women.
The two main aspects that influence result-stability are the nature and number of individuals and diseases that are included in the analysis 
. The diagnoses in this study were taken from the electronic medical records of 275,682 primary care adult patients in the context of a national health system with universal coverage, which has been shown to lead to more reliable and representative conclusions compared to those derived from survey-based studies 
Regarding the statistical methodology used, the exploratory factor analysis applied in this study is the preferred method when the objective is to explore statistically significant stable disease clusters 
. In the absence of inferential statistics on which to base the analytical choices related to this type of method, the decisions were based on a set of rules and recommendations for the social sciences that have previously been described by Costello et al. 
. Thus, this study adhered to the following criteria: (a) the use of the principal factor method as the technique for extracting factors by assuming a nonparametric distribution of binary data (presence/absence of disease); (b) the use of scree plots together with a clinical assessment of the results to select the number of factors; (c) the oblique rotation of factors; (d) factor scores greater than 0.30, which is set as the minimum acceptable value for the correlation to be significant from a clinical and statistical perspective; (e) the detection of few or no diseases with strong associations with several factors; (f) the identification of at least three diseases per factor; (g) obtaining samples from more than 100 individuals and (h) the presentation of the factor scores.
It is noteworthy that this study met most of these standards. The decision to reduce the minimum factor score to 0.25 (or even 0.20) was due to the expectation that there would be a significant number of associations among diseases due to chance (i.e., concurrent multimorbidity). Therefore, a more permissive threshold was established. On the other hand, despite the fact that the depressive pattern comprised only two diseases (depression and behavioural disorders, such as insomnia), this pattern has been previously identified in the literature 
. Moreover, the goodness-of-fit values for the models, expressed as the percentage of the accumulated variance (i.e., between 14.80% and 26.87%) and the sampling adequacy (i.e., KMO measures between 0.50 and 0.71) were above the acceptable lower limits.
Although several hypothesis have been put forward on the pathophysiological processes underlying the five multimorbidity patterns brought to light in this study, the former must be interpreted with the necessary caution since the study design (i.e. transversal) does not allow to establish the sequence in which diseases cluster within a pattern. Longitudinal studies would be required to corroborate the suggested causal associations as well as to help elucidate those disease associations that could not be explained in the present study.
As stated earlier, this study was based on information that was recorded in the primary care electronic medical record system during medical visits. Although there are many benefits of this methodology, it can also limit the data. The workload of healthcare professionals and the structure of the applied diagnostic coding system (i.e. ICPC) often lead certain information regarding disease history not to be recorded in the individual patient medical histories. Therefore, the frequencies of many diseases are often underestimated. We suspect that this may have occurred for smoking, which is systematically underreported in a population with a very high prevalence of smokers 
. This led us to exclude this information from the analysis and therefore prevented us from providing a plausible explanation to some of the associations found with the cardio-metabolic pattern. However, the fact that we focused this study on chronic diseases, which have diagnostic codes that remain in the health record over time, may have helped to minimise this problem. Furthermore, this constraint is expected to be minimal, given the quality criteria that were used to select the centres included in the study.
A potential overrepresentation of certain diagnoses may exist when these are associated with other diseases for which treatment protocols recommend periodic health reviews. In other words, the higher frequency of visits by patients with these diseases would affect the likelihood that associated diseases were diagnosed (i.e., observer bias). To avoid this bias, future studies should be based on the entire population rather than solely on those users of the primary care services.
It is important to note that this work focused on chronic diseases. Although this may ensure comparability with other large studies across different international contexts, future research should consider including the whole range of diseases seen in patients. This would be especially true where the boundaries between “chronic” and “acute” diseases are not always clear, as was demonstrated by Starfield 
. Often, the recurrence of acute diseases causes them to be chronic, and they should be regarded as chronic diseases in all of their dimensions 
Although the different age groups were defined according to the expected biological homogeneity of individuals among groups, the selected thresholds could have influenced the nature of the obtained multimorbidity patterns. The use of different and/or narrower age groups may be advisable in future studies.
To conclude with the potential limitations of this study, it should be noted that this study focused on the occurrence and the simultaneous association of diseases that were defined and registered from the perspective of the professionals caring for the patient and not by the patient him/herself. It is possible that many of the disease manifestations affecting the quality of life of the patient are not met by the defined criteria. Understanding the diseases from this perspective would require complementary methodological approaches that are beyond the scope of this work.
Comparison with other studies
It is difficult to compare these results with those of other studies because of differences in the disease inclusion criteria, the study populations and the data sources used. This study was also the first to extend the analysis to age groups under 65. The study by Schäfer et al. 
, which was also based on a factor analysis of clinical-administrative data for the German primary care population over 65 years of age, identified three multimorbidity patterns for both men and women: 1) cardiovascular and metabolic disease-related, 2) anxiety, depression, somatoform disorders and pain-related and 3) neuropsychiatric disorder-related. Palomo et al. evaluated the associations between hypothyroidism and various diseases in the Spanish population treated in primary care 
, and these associations were found to precisely conform to the thyroid pattern noted in this study. In the study by Holden et al. based on surveys of Australian workers 
, it was concluded that different health problems are grouped beyond their physiological link with an organ or system. Britt et al. described multimorbidity patterns in the Australian population attended by primary care physicians 
using the Cumulative Illness Rating Scale (i.e., the CIRS) for the categorisation of clinical entities. Marengoni et al. used a different statistical technique (i.e., cluster analysis) to analyse clusters of diseases in people older than 76 years of age 
. This technique was also used in the study by Cornell et al. 
, which identified six patterns of multimorbidity among veterans of the United States. These referenced results largely coincide with those from this study and support the existence of the following multimorbidity patterns: mechanical obesity-related, metabolic, neurovascular, liver disease-related, dual diagnosis-driven (psychiatric-substance abuse); and anxiety and depression-related.
Implications for health systems
Despite the significance of multimorbidity, especially in older individuals, the scientific evidence provided by the clinical practice guidelines is not appropriate for patients with multiple diseases 
, and there is a clear gap in the design of specific intervention programmes for the management of these patients 
. One of the main consequences of multimorbidity is polypharmacy, which entails the consequent risk for drug interactions and side effects 
. Accordingly, this study observed examples of these phenomena, including the possible relationship between chronic liver disease and the treatment of ischemic heart disease with statins in men aged 45–64 
and the possible interaction between the use of antipsychotics for the behavioural alterations of dementia and the onset of Parkinsonism after the age of 65 
From the standpoint of healthcare management, strong and developed primary care is the best solution to the complex needs of patients with multimorbidity 
. However, the fragmentation of the current healthcare model hinders integrated and coordinated care between the primary and specialised care levels and between the healthcare and social sectors, which leads to inefficiencies in the care of patients with multimorbidity 
Moreover, the importance of other approaches beyond those based on health services, which are needed to ensure the health of the population, should be stressed. For example, the promotion of a healthy lifestyle is closely linked to a decline in the incidence of obesity 
, and as this study has shown, obesity was present in three of the patterns described (cardio-metabolic, mechanical-obesity-thyroidal and psychiatric-substance abuse).
Finally, a change in the focus of etiologic research is required for the study of diseases and their relationship to joint presentations, and an analysis of underlying factors, including pathophysiological, genetic, iatrogenic and/or socioeconomic factors should be promoted.
Conclusion and future research
The results of this work have demonstrated that the existence of non-random associations between chronic diseases is a reality for the entire population, not only the elderly.
These disease associations give rise to clinically consistent multimorbidity patterns that affect a significant proportion of the population. Most importantly, when studying the lifetime disease process, there are underlying pathophysiological phenomena upon which action can be taken both from a clinical, individual-level perspective and from a public health or population-level perspective.
Knowledge regarding the main disease patterns and their evolution with age should facilitate the design of tools, such as clinical practice guidelines, treatment protocols for patients with multimorbidity and healthcare information systems with alarm signals to monitor for the duplication of visits and diagnostic tests, the incidence of polypharmacy and potential adverse drug reactions, inappropriate hospitalisations and even mortality.
All information available in the health systems should be properly managed to maximise the individual attention given to these patients and to ensure the efficient use of resources. This will positively impact the overall health of the population.
The integration of information is a basic strategy that should be accompanied by structural and functional measures that promote coordination between different actors and levels of the care and that ensure safe and adequate communication among the professionals who care for such patients.
Multimorbidity is a public health problem that must be considered by the planning and organisational frameworks of the health system. This work aimed to provide clinically relevant information to improve this process.