How Hidden Markov Models Could Elucidate Multimorbidity in Aging

In a new study, researchers investigated longitudinal multimorbidity patterns among older adults from a Swedish urban population.

Figure 1. Evolution and transitions of multimorbidity patterns over time by age group (N=3,363).
Figure 1. Evolution and transitions of multimorbidity patterns over time by age group (N=3,363).
Listen to an audio version of this article

Multimorbidity is a term that refers to living with two or more chronic diseases at the same time, and the prevalence of this phenomenon increases with age. In addition, humans tend to evolve and transition into distinct patterns of multimorbidity. These still ill-defined patterns of multimorbidity may offer a window of opportunity for researchers. Since the aging population continues to grow in many parts of the world, researchers are motivated to better understand these patterns and how they evolve and transition over time in order to develop interventions and therapeutics for healthier aging. However, this is a challenging task for several reasons.

“Multimorbidity is associated with a higher risk of polypharmacy and decreased quality of life, and challenges the decision-making of clinicians that lack effective guidelines for the management and treatment of patients with cohexisting complex diseases [4].”

Multimorbidity Patterns

While researchers have investigated multimorbidity, not all studies are created equal—rendering meta-analyses largely incongruent (thus far). One reason the evolution of multimorbidity patterns is so challenging to study is because most study designs are not powered to account for the dynamic nature of multimorbidity in old age. Another reason is that various studies use different lists of diseases. (Some studies include ten conditions or less and others include 200+ conditions.) Finally, most statistical methods used to organize data are not able to properly handle the complexity of multimorbidity.

“Exploring how multimorbidity patterns evolve throughout people’s lives and the time subjects remain within specific patterns is still an under-researched area [7, 8]. The understanding of how diseases cluster longitudinally in specific age groups would pave the way to the design of new prognostic tools, as well as new preventive and, eventually, therapeutic approaches.”

In a new study, researchers Albert Roso-Llorach, Davide L. Vetrano, Caterina Trevisan, Sergio Fernández, Marina Guisado-Clavero, Lucía A. Carrasco-Ribelles, Laura Fratiglioni, Concepción Violán, and Amaia Calderón-Larrañaga from the Jordi Gol i Gurina University Institute Foundation for Research in Primary Health Care (IDIAPJGol), Universitat Autònoma de Barcelona, Karolinska Institutet, Stockholm University, Stockholm Gerontology Research Center, University of Ferrara, Madrid Health Service, and Universitat Politecnica de Catalunya investigated the evolution of multimorbidity patterns in a longitudinal study using complex statistical models. The team published their research paper in Aging’s Volume 14, Issue 24, entitled, “12-year evolution of multimorbidity patterns among older adults based on Hidden Markov Models.”

Hidden Markov Models 

“Recently, several advanced machine-learning techniques such as non-hierarchical and hierarchical clustering techniques have been used to explore multimorbidity patterns.”

Hidden Markov Models (HMM) were developed based on the Bayesian Information Criterion. The Bayesian Information Criterion is an algorithm of inference that is used to select the best model from a set of possible models. It is a powerful technique for analyzing temporal data that can capture dynamic changes in longitudinal patterns over time. Since HMMs can account for complex longitudinal data, they are well-suited to investigate the dynamics of multimorbidity over time.

The Study

In this study, HMMs were used to investigate 3,363 older adults from the Swedish National study on Aging and Care in Kungsholmen (SNAC-K) and the evolution of their multimorbidity patterns over the course of 12 years. The aim of this research was to explore the evolution of these patterns across decades of life in older adults and to examine how they transition across different chronic diseases when further chronic diseases arise. In this cohort of study participants, the average age was 76.1 years old, 66.6% were female and 87.2% had multimorbidity at baseline.

The researchers divided the participants into three groups based on age: sexagenarians (between 60 and 66 years), septuagenarians (between 72 and 78 years) and octogenarians (81 years and over). Data used in the HMMs included age, gender, education level, self-reported chronic diseases and medications, results from a Mini Mental State Examination (MMSE), and walking speed. Data were collected from participants at baseline and at six and 12 years (three separate time points).

“At each follow-up wave, SNAC-K participants undergo an approximately five-hour-long comprehensive clinical and functional assessment carried out by trained physicians, nurses, and neuropsychologists.”


The team identified four longitudinal multimorbidity patterns in each decade. The Unspecific pattern consists of participants with no specific pattern of multimorbidity. In all decades, participants showed the shortest permanence time in the Unspecific pattern. The researchers also included categories for participants who dropped out of the study or passed away.

Next, the top 10 diseases were selected out of each age group at each follow-up wave to identify the most common multimorbidity patterns. Among the sexagenarians, the multimorbidity patterns were clustered into cardiovascular and anemia, cardio-metabolic, and psychiatric-endocrine and sensorial. Among the septuagenarians, the multimorbidity patterns were clustered into cardiovascular and diabetes, neuro-vascular and skin-sensorial, and neuro-psychiatric and sensorial. Among the octogenarians, the multimorbidity patterns were clustered into respiratory-circulatory and skin, cardio-respiratory and neurological, and neuro-sensorial. The data showed that participants commonly shifted from one pattern to another. (See Figure 1.)

“In this study we identified and characterized longitudinal multimorbidity patterns among older adults from a Swedish urban population, and estimated the time they spent in each pattern as well as the probability of transitioning across different patterns throughout a 12-year follow-up period.”


“Our statistical approach enabled us to model the evolution and transitions of multimorbidity over time, and the results of this could be applied in the interests of healthier aging. Moreover, the age-stratified analyses allowed us to identify which disease combinations and transitions were more prevalent in each decade.”

The findings of this study suggest that multimorbidity patterns change with age and highlight the importance of understanding the dynamic nature of multimorbidity over time. Through the use of HMMs, this research was able to detect changes in the prevalence and transition of multimorbidity patterns across different decades of life. These findings can help healthcare providers and researchers better understand the complex nature of multimorbidity and develop more effective interventions for older adults. Furthermore, this research provides evidence that the use of HMMs to study longitudinal data is a useful tool for further research into multimorbidity. Additional studies with more data is needed to gain a better understanding of the interplay between multimorbidity and aging.

“Our study provides evidence that multimorbidity is dynamic and heterogeneous in old age. With increasing age, older adults experience decreasing clinical stability and progressively shorter permanence time within one same multimorbidity pattern. Moreover, a significant proportion ranging between 5.9%-22.6% belongs to an Unspecific pattern with a low burden of diseases and a promising preventive potential. Adding new variables related to drug use, environmental and genetic factors, and/or frailty to the longitudinal analysis of multimorbidity patterns may allow optimizing the epidemiological understanding and applicability of these models for patient-tailored prevention and management strategies.”

Click here to read the full research paper published by Aging.

AGING (AGING-US) VIDEOS: YouTube | LabTube |

Aging is an open-access journal that publishes research papers bi-monthly in all fields of aging research. These papers are available at no cost to readers on Open-access journals have the power to benefit humanity from the inside out by rapidly disseminating information that may be freely shared with researchers, colleagues, family, and friends around the world.

For media inquiries, please contact [email protected].

Trending With Impact: Machine Learning Predicts Human Aging

Machine learning and a broad range of biochemical and physiological traits were used to develop a new composite metric as a potential proxy for an underlying whole-body aging mechanism.


The Trending With Impact series highlights Aging (Aging-US) publications that attract higher visibility among readers around the world online, in the news, and on social media—beyond normal readership levels. Look for future science news about the latest trending publications here, and at

Listen to an audio version of this article

Will you age quickly or slowly? Is it possible to predict how long you will live based on your genetics, lifestyle and other traits? In a new study, a team of researchers—from the National Institutes of Health’s National Institute on Aging, University of California San Diego, University of Michigan, Consiglio Nazionale delle Ricerche, Azienda Sanitaria di Firenze, and ViQi, Inc.—sought to answer these questions by developing a novel framework designed to estimate human physiological age and aging rate. Their trending paper was published by Aging (Aging-US) in October 2021, and entitled, “Predicting physiological aging rates from a range of quantitative traits using machine learning”.

“We present machine learning as a promising framework for measuring physiological age from broad-ranging physiological, cognitive, and molecular traits.”

Machine Learning

Machine learning is an important development in computer science that uses artificial intelligence. Algorithms and data (figured and input by human intelligence) are programed to automatically learn and improve through experience and new data. Machine learning approaches allow researchers to build mathematical models onto training data to predict target variables—target variables including human physiological age and rate of aging.

“Here we use a machine learning approach with a broad range of biochemical and physiological traits including blood phenotypes (e.g., high-density lipoprotein), cardiovascular functions (e.g., pulse wave velocity) and psychological traits (e.g., neuroticism) as main groups from the SardiNIA longitudinal study of aging [48, 49] to estimate human physiological age, a metric for phenotypic and functional age progression [7].”

Subjects and Traits

Two very interesting study populations were included in this particular aging model. People living in Sardinia—an island off the coast of Italy and one of the first identified “Blue Zones”—are well-known for their long lives. They are currently contributing to a large longitudinal study on human aging, known as the SardiNIA Project. Data from the SardiNIA Project was used to develop the aging model in the current study. 

“Funded by the National Institute on Aging in 2001, the SardiNIA Project (age range 14.0 to 101.3 years, with a mean of 43.7 years; 57% female) is a longitudinal study of human aging on the island of Sardinia, which is notable for its long-lived population [48, 49].”

The second cohort included in the current study was collected from the InCHIANTI study. Participants in this longitudinal population-based study were predominantly older adults living in Tuscany, Italy. After collecting the initial datasets from both cohorts, the researchers reduced the datasets using a “cleaning” strategy they developed. After cleaning, the number of subjects in the study went from 6165 to 4817, and the number of traits included in the algorithms went from 183 to 148. The researchers then configured the selected subjects and traits using computational algorithms and machine learning. Traits were ranked based on importance and weighted accordingly using algorithms the researchers developed. Study methods and materials were detailed thoroughly in the paper and its supplemental materials.

Supplementary Figure 1. Computational workflow for measuring physiological age and physiological aging rates (PAR) using the machine learning framework.
Supplementary Figure 1. Computational workflow for measuring physiological age and physiological aging rates (PAR) using the machine learning framework.


The team developed a promising new composite metric and was able to closely predict chronological age using their machine learning strategy. After they effectively estimated physiological age and validated their results, the researchers then used the ratio of physiological and chronological age to determine physiological aging rate, or PAR. Interestingly, the researchers observed that PAR was highly correlated with the epigenetic aging rate (EAR), which is a DNA methylation-based measure of aging. In addition, the researchers demonstrated that individuals with lower PARs outlived individuals with higher PARs. PAR may be a new proxy for an underlying whole-body aging mechanism.

“The efficacy of treatments aimed at slowing the aging process has traditionally been evaluated using individual biomarkers or limited collections of related biomarkers. Our current study has shown that PAR is a significant predictor for survival and correlated with epigenetic aging rate, providing evidence for a good measurement of ‘aging’.”

Click here to read the full research paper published by Aging (Aging-US).


Aging (Aging-US) is an open-access journal that publishes research papers monthly in all fields of aging research and other topics. These papers are available to read at no cost to readers on Open-access journals offer information that has the potential to benefit our societies from the inside out and may be shared with friends, neighbors, colleagues, and other researchers, far and wide.

For media inquiries, please contact [email protected].

Deep Learning Technology Consolidates Wearable Sensor Data

Smart watch / Smartphone

The Top-Performer series highlights papers published by Aging that have generated a high Altmetric attention score. Altmetric scores, located at the top-left of trending Aging papers, provide an at-a-glance indication of the volume and type of online attention the research has received.

Read Aging’s Top 100 Altmetric papers.

Listen to an audio version of this article

Wearable sensors (smartwatches, smartphones, and other devices) allow users to monitor some biomarkers of their own health with mobile biofeedback technology. In 2019, one-in-five adults in the United States reported regularly using a wearable fitness tracker or smartwatch. Since the COVID-19 pandemic, mobile downloads of health and home fitness apps have increased by 46%—in addition to a boom in wearable sensor use.

“Wearable device motion data have already been used for monitoring acute illnesses including detection of early signs of the outbreak of influenza-like illnesses [28] and COVID-19 [3034].” 

Large quantities of these data are being collected consistently from individual users. This potentially useful information is also being collected from large populations of people living in different countries, working in different occupations, with unique health statuses, and across multiple environmental seasons and stages of life. Wearable sensor data provides an opportunity to conduct large-scale studies that could lead to new global discoveries in aging and disease research.

“In fact, only mobile technology can support large-scale studies involving monitoring of early signs of a disease or measuring recovery rates, all requiring sampling more often than once per week.”

However, there are a number of different manufacturers of wearable sensors, smartwatches, and mobile devices. In addition to the inevitable inaccuracies, such as missing data, outliers, and even seasonal variation of physical activity, there are also varying measurements between devices of different manufacturers. These inaccuracies and variations create inconsistencies when comparing large-scale data from wearable sensors.

“We applied deep learning technology to systematically address these challenges.”

In 2021, researchers from Singapore’s Gero AI and Russia’s Moscow Institute of Physics and Technology authored a paper, published in Aging’s Volume 13, Issue 6, and entitled, “Deep longitudinal phenotyping of wearable sensor data reveals independent markers of longevity, stress, and resilience.” To date, this top-performing research paper has generated an Altmetric attention score of 43

The Study

“We trained and characterized a simple model that learns physical activity patterns from wearable devices, which are directly associated with morbidity risks on the population level.”

Three wearable sensor manufacturers were assessed in this study: UK Biobank, NHANES, and Healthkit. Researchers collected wearable sensor data for physical activity (steps per minute) from 103,830 users over the course of one week and, among 2,599 users, up to two years of data were collected. The team trained and validated a deep learning neural network technology—the GeroSense Biological Age Acceleration (BAA) system—to extract health-associated features from the physical activity recordings.  

“GeroSense BAA model employs additional neural network components to address this domain shift problem to ensure learning device-independent representations of the input signal.”


“We demonstrate that deep neural networks trained to predict morbidity risk from wearable sensor data can provide a high-quality and cheap alternative for BAA determination.”

The researchers explained that the application and wide deployment of their GeroSense BAA system may provide the means to accurately monitor stress and resilience in response to environmental conditions and interventions among people in different populations, countries, and socio-economic groups. 

“We hope that future developments will lead to further applications of AI in geroscience research, public health, and policy decision-making.”

Click here to read the full research paper, published by Aging.


Aging is an open-access journal that publishes high-quality research papers bi-monthly in all fields of aging research and other topics. These papers are available to read at no cost to readers on Open-access journals offer information that has the potential to benefit our communities from the inside out and may be shared with friends, neighbors, colleagues, and other researchers, far and wide.

For media inquiries, please contact [email protected].

  • Follow Us