Relationships between Charlson comorbidity index associated biomarkers and outcomes among participants in the Malmö diet and cancer study

Abstract Introduction The aim was to evaluate two biomarker scores trained to identify comorbidity burden in the prediction of specified chronic morbidities, and mortality in the general population. Methods Cardiovascular biomarkers were measured in the cardiovascular cohort of the Malmö Diet and Cancer Study. A score of 19 biomarkers associated with Charlson Comorbidity Index (CCI) was created (BSMDC). Individuals with CCI diagnoses and other major comorbidities were excluded. Another score of 11 biomarkers associated with comorbidity burden from a previous study of acute dyspnea was also created (BSADYS). The scores were prospectively evaluated for prediction of mortality, and some chronic diseases, using Cox Proportional Hazards Model. Results Fully adjusted models showed that BSMDC was significantly associated per 1 SD increment of the score with incident COPD, 55%, and congestive heart failure, 32%; and with mortality, 33% cardiovascular, 91% respiratory, 30% cancer, and 45% with all-cause mortality. The BSADYS showed no association with these outcomes, after simultaneous inclusion of both biomarker scores to all the clinical covariates. Conclusion BSMDC shows strong prediction of morbidity and mortality in individuals free from comorbidities at baseline, and the results suggest that healthy individuals with high level of BSMDC would benefit from intense preventive actions. Key Messages A score of 19 biomarkers associated with Charlson Comorbidity Index was created, the Biomarker Score of Malmö Diet and Cancer study (BSMDC). The created BSMDC index was associated with both incident COPD, and incident CHF. BSMDC was also associated with cardiovascular mortality, respiratory mortality, cancer mortality and with all-cause mortality.


Background
Biomarkers are increasingly being considered to improve risk stratification and thereby guide clinical decisions, both in the acutely ill patients presenting at the emergency department and in the general population attending general practitioners.Whereas risk prediction and prevention and treatment of atherosclerotic cardiovascular disease has resulted in substantial reductions of coronary artery disease and survival after myocardial infarction, the burdens of congestive heart failure (chF) and chronic Obstructive Pulmonary Disease (cOPD) are still enormous (Braunwald 1997;halpin et al. 2019).smoking is a strong risk factor for both chF and cOPD, but other shared pathophysiology patterns between the two diseases is present also in non-smokers (engström et al. 2010).
Multimorbidity is increasing with increasing age and is an important factor for mortality (Forslund et al. 2021).Multimorbidity could be assessed in different ways, such as the charlson index (charlson et al. 1987).however, finding Biomarkers; charlson comorbidity index; cOPD; congestive heart failure; mortality; acute dyspnea; Proteomics; the Malmö Diet and cancer study; emergency department; biomarker; aDYs biomarkers with strong association to a multimorbidity index would be of great value.
in a previous aDYs study on patients with acute dyspnea, where chF and cOPD were the dominating underlying diseases (Wessman et al. 2021), it was found that a score consisting of the sum of 11 weighted and standardized values of serum protein biomarkers, each associated with comorbidity burden was superior in predicting mortality as compared to actual comorbidity burden. in this previous study, performed on 774 patients with acute dyspnea, it was found a strong and significant relationship between the comorbidity-associated biomarker score and all-cause mortality.
in the current study we switched focus from the setting of acute dyspnea to the general population, and investigated middle-aged individuals included in the Malmö Diet and cancer study, the cardiovascular cohort (MDc-cc).We hypothesized that biomarkers with a strong association with charlson comorbidity index in the population (cci) (charlson et al. 1987), as well as biomarkers with a strong association with multimorbidity burden, as defined in a previous aDYs study (Wessman et al. 2021) on acute dyspneic patients, may aid in identification of healthy individuals at risk of new-onset chF, new-onset cOPD and premature dyspneic mortality as well as all-cause mortality among subjects without such comorbidities at baseline.to address these two issues we investigated partly if the same comorbidity-associated biomarkers summed up into a score (BsaDYs), and partly if a new comorbidity biomarker score associated with the presence of a positive cci at inclusion in the population based MDc-cc (BsMDc), predicted new onset of chF and new-onset cOPD, which are the two most common dyspnea related diseases, as well as deaths in cardiovascular, respiratory and malignant diseases and all-cause mortality in healthy middle-aged individuals during long-term follow-up in the MDc-cc.

Study population
The MDC study the Malmö Diet and cancer study (MDc) started in the early 1990s on the initiative of the swedish cancer society (Manjer et al. 2001).the original aim was to study the relationship between diet and risk of developing cancer in the population.the study is a prospective population-based cohort study involving 30 446 randomly selected men (born 1923-1945) and women (born 1923-1950).the subjects in the study underwent collection of blood samples at baseline between 1991 and 1996.the blood samples were frozen and saved in a biobank and later examined with biochemical and genetic analyses.a sub cohort of 6103 of these individuals (Figure 1) was included in a cardiovascular disease study (the cardiovascular cohort of MDc, MDc-cc), to study the relationship between ultrasonically estimated carotid atherosclerosis and the risk of developing cardiovascular diseases in the population.the baseline survey consisted of a questionnaire concerning lifestyle, socio-economic factors, smoking, medications, and previous diseases and illnesses.the body constitution was examined, and height, weight, hip, and waist circumference were measured.their fat percentage was calculated with a bioimpedance measurement.all blood samples were separated and stored at either minus 80° c or minus 140° c. all participants provided written informed consent.study protocols were approved by the ethical committee at lund University, lund, sweden.the MDc baseline examination included a questionnaire which included smoking habits.smokers were registered as either ever smokers or never smokers.information about prior diseases was achieved through record linkage. in MDc there are information on the presence of diseases and comorbidities, based on the swedish National inpatient registry (ludvigsson et al. 2011) which was used to define by charlson comorbidity index (cci) (charlson et al. 1987).the cci is a weighted comorbidity index, which gives a total score of weighed points according to the total number and severity of diseases.in the MDc-cc 4% of the individuals had at least one registered cci disease.
serum samples were analyzed for biomarkers using Olink Proseek® Multiplex cardiovascular i 96X96 kit (http://www.olink.com/) at scilife laboratories in Uppsala, sweden in the MDc-cc participants.the panel consists of 92 biomarkers and is a proximity extension assay (Pea) that measures the relative abundance of circulating proteins.the panel contains known human cardiovascular and inflammatory markers as well as some exploratory human proteins with great potential as new cVD markers (lind et al. 2021).each analyte in the panel has according to Olink Proteomics aB been assessed in terms of sample material, specificity, precision, sensitivity, dynamic range, matrix effects and interference.
Deciding on a minimum of at least 4 500 analysis results, this left us with complete analyses results on 82 Olink biomarkers for n = 4 531 individuals of the Olink cVD i panel.NtproBNP was one of the biomarkers with missing data in the Olink panel.as NtproBNP is an important biomarker among other things for validating dyspnea and severity of heart failure, we wanted to include NtproBNP in our study.instead NtproBNP was separately analyzed for the same individuals at Dade Behring holdings, inc.(joined with the existing business of siemens healthcare Diagnostics in 2007) with "the stratus® cs acute care™ Nt-proBNP assay.With NtproBNP included we thus had complete biomarker results on 83 biomarkers in n = 4 531 individuals.

Creation of the BSADYS and BSMDC scores
For BsMDc we first tested associations between the 83 biomarkers (simultaneously included and additionally adjusted for age and sex using linear regression with a stepwise selection method, with cci (presence of at least one cci comorbidity) as the outcome for 4531 individuals (table s1).twenty biomarkers had a significant association to cci.We log-transformed and standardized these 20 biomarkers and made a new linear regression to determine each biomarker's beta-coefficient vs presence of cci. in this second step only 19 biomarkers had a p-value< 0.05 (table 1).We thus created a summed-up score of these 19 standardized biomarkers, with each biomarker individually weighted for its beta-coefficient.this summed-up score was then standardized (again expressed as number of standard deviations from the mean).this standardized score we refer to as the BsMDc (table 2).
For BsaDYs, we used the same 11 biomarkers as in a previous study on the aDYs cohort (Wessman et al. 2021) (table 2).With the 11 comorbidity-associated biomarkers as a basis, the log-transformed values of these biomarkers from the MDc-cc individuals were weighted, based on their beta-coefficients in relation to the comorbidity score in aDYs, and put on a standardized scale (expressed as number of standard deviations from the mean), creating a new standardized biomarker score for these 11 biomarkers which we refer to as the BsaDYs.We only wanted to include

Endpoints
Our endpoints were new-onset of cOPD, new-onset of chF, deaths due to cVD, respiratory diseases, cancer as well as all-cause mortality.endpoints were ascertained by linkage of swedish 10-digit personal identification numbers to national swedish registers (swedish hospital inpatient Registry, swedish National cause of Death Registry) maintained by the swedish National Board of health and Welfare. the onset of chF was retrieved through record linkages with the swedish hospital inpatient Registry for the international classification of Diseases (icD) 8 th , 9 th and 10 th revision using the following diagnosis codes for a primary hospital diagnosis: for icD-8 427.00, 427.10, and 428.99; for icD-9 428; and for icD-10 i50 and i11.0.an earlier validation study of these registers has been previously found that a diagnosis of heart failure exhibits a high case validity (ingelsson et al. 2005).information regarding the diagnosis of new-onset of cOPD according to icD-8 (490-492), icD-9 (491-492, 496) and icD-10 (J41-J44), was also retrieved from the swedish hospital inpatient Registry.information on death causes and all-cause mortality was retrieved from the swedish National cause of Death Registry (i.e.icD-10 chapter "i" for Diseases of the circulatory system, icD-10 chapter "J" for Diseases of the respiratory system, icD-10 "c00-D48" for Neoplasms or corresponding in icD-9).the swedish hospital inpatient Registry and the swedish National cause of Death Registry) have both previously been validated and described and include both inpatient and outpatient specialized care but does not give information from the Primary health care facilities (inghammar et al. 2012;Brooke et al. 2017).

Follow-up
Data concerning new onset of cOPD from the swedish hospital inpatient Registry, extended until 2016-12-31, with a mean follow-up of 21.4 ± 4.9 years.concerning new-onset chF, the data from the swedish hospital inpatient Registry the end of follow-up was 2018-12-31, with a mean follow-up of 23.0 ± 5.4 years.Mortality data from the National swedish Death Registry extended until 2018-12-31, with a mean follow-up of 23.2 ± 5.1 years.

Statistics
Data are presented as median (interquartile range) or mean (± sD), depending on presence or absence of normal distribution of data.Group-wise differences of continuous variables were compared using aNOVa or Kruskal Wallis test when appropriate.categorical variables were compared between groups using chi-2 test.to create BsMDc, we performed stepwise linear regressions with cci as dependent variable adjusted for age and gender, and the 83 biomarkers entered as independent variables (with stepwise exclusion).
We used cox Proportional hazards model to relate exposures to variables in relation to our outcomes after exclusion of individuals with history of any of the cci comorbidities at baseline and individuals who did not have complete data on all biomarkers included in the two scores or data on any of the covariates adjusted for, leaving us with n = 3616 middle-aged individuals with complete data (Figure 1).We adjusted for age, sex, ever smoking, diabetes, antihypertensive medication, systolic blood pressure, body mass index and history of coronary artery disease. in a second step, both biomarker scores (BsaDYs and BsMDc) were entered simultaneously in the model together with all the above covariates.a 2-tailed significance level of (p < 0.05) was considered statistically significant.ROc analyses were performed, and aUc values were reported for the BsMDc score in relation to the outcomes.all calculations were done with iBM sPss statistics, version 27.

Baseline characteristics
the mean age at inclusion for our cohort was 57.4 ± 5.9 years (table 3) with a majority of females (63.5%).the systolic blood pressure and the BMi were fairly normal.Of the individuals 57.3% were ever smokers (present or previous), and 15% were on antihypertensive medication.

Outcome analyses
During follow up of the 3616 individuals who at baseline were free from cci defined disease as well as caD and DM not resulting in a cci, there were 331 cases of incident cOPD, and 225 cases of incident chF.there were 1266 cases of all-cause death, whereof 353 deaths were due to cVD, 86 due to respiratory diseases and 472 due to malignant diseases.in multivariate cox regression models for each biomarker score separately, adjusted for age, gender, and all covariates (antihypertensive medication, systolic blood pressure, ever smoking, body mass index) the BsaDYs was significantly associated with an increased risk of incident cOPD, incident chF, all-cause mortality, cVD mortality and respiratory mortality, but no significant association with death due to cancer (tables 4,5).the BsMDc was significantly associated with incident cOPD, incident chF, all-cause mortality, cVD mortality, respiratory mortality, and cancer mortality (tables 4,5).ROc curves and aUc values are presented for BsMDc in relation to incident cOPD, incident chF, all-cause mortality, cVD mortality, respiratory mortality, and cancer mortality (Figures 2-7).
after simultaneous inclusion of both biomarker scores on top of all the clinical covariates, the BsMDc was significantly associated with all endpoints, i.e. cOPD, chF, all-cause mortality, cVD mortality, respiratory mortality and cancer mortality, whereas the BsaDYs was no longer associated with any of the endpoints (tables 4,5)

Discussion
the main finding of the study was that the BsMDc, i.e. the biomarker score associated with cci registered at inclusion, showed strong, significant, and independent association with new-onset chF, new-onset cOPD, deaths in cVD, deaths in respiratory diseases, deaths in cancer as well as all-cause mortality among individuals free from cci and other major comorbidities at baseline.interestingly, even if the effect sizes of the BsMDc vs endpoints were generally greater than those of the BsaDYs (Wessman et al. 2021), both scores individually predicted incident chF, incident cOPD, deaths in cVD and respiratory diseases as well as all-cause mortality.although this finding needs replication, it suggests that for prediction of chF, cOPD, deaths in cVD and respiratory diseases as well as all-cause mortality, the two scores may give complementary information in addition to clinical risk factors.
the fact that certain biomarkers that predict poor outcome in acutely ill patients also may be predictive for poor outcome in a healthy population, is logical since several such previous examples exist.For example, cRP which is used to diagnose acute infections, and troponin-t, which is used to predict acute myocardial infarction, are gaining increasing interest as markers for poor outcome (i.e.mortality and cardiovascular disease incidence) at the population level (Ridker and silvertown 2008;avan et al. 2018;Jia et al. 2019;terho et al. 2019).there are also studies on biomarkers predicting risk for cardiovascular disease in an apparently healthy population (esteghamati et al. 2014;Kouvari et al. 2019).Many studies are carried out studying one individual biomarker at a time in relation to a specific individual disease, both to diagnose but also to predict progression, prognosis, etc. to our knowledge there are no previous studies on how either a score of biomarkers associated with comorbidity burden in an emergency setting or a score of biomarkers associated with cci in a middle-aged normal population, can predict future development of dyspnoeic diseases and premature deaths. in this study we hypothesized that the aggregated information of a broad set of circulating protein biomarkers in blood, would serve as a biological fingerprint of the complex situation of both chronic and more acute processes, as well as various diseases and multimorbidities.We chose to study new-onset of chF and new-onset cOPD as these are the two main dyspneic diseases.importantly, even if the BsMDc was derived from relationships with cci, its prognostic information is completely independent from cci as subjects with any cci comorbidity at baseline were excluded in the outcome analyses.thus, the BsMDc possibly represents a biological fingerprint of having undiagnosed or a potential of developing cOPD and chF, as well as predicting deaths in cVD, deaths in respiratory diseases, deaths in cancer as well as with all-cause mortality many years in advance.Moreover, as we did not train the BsMDc to identify incident disease and death, but rather to identify individuals with cci, model over-fitting is unlikely to be a major problem.still, it will be important to replicate the relationships between the BsMDc and the six different outcomes in independent cohorts.
since many of the biomarkers included in the BsMDc are associated with cardiovascular diseases, it is interesting that the BsMDc score had a significantly stronger association with new-onset cOPD than with new-onset chF, as swell as a significantly stronger association with deaths in respiratory diseases than with deaths in cardiovascular diseases.One possibility is that susceptibility to cOPD causes cardiovascular stress and elevation of such biomarkers.the relationship between cOPD and chF is well established (Macchia et al.    2012; Müllerova et al. 2013).however, it does not explain why the relationship is weaker for chF.cOPD shares the same etiology and epidemiology as chF to a rather high extent, why we would have expected more or less the same level of significance for both cOPD and chF.But despite their sensitivity to the heart, the markers are not as strongly associated with chF as with cOPD.there are a fairly large number of ever-smokers in the cohort (57%).even if the results are adjusted for smoking, could that still be part of the reason?this remains speculative though, and additional in particular mechanistic studies are needed.
BMi is also a factor of interest.low BMi or underweight is associated with increased mortality (Wändell et al. 2009)), and especially in respiratory diseases (Global et al. 2016).however, the obesity paradox, i.e. a similar or even lower mortality in some groups especially among elderly, could also be a factor of importance in the present study (Wändell et al. 2021).
still, the weaker effects related to the BsaDYs compared to BsMDc is probably a consequence of that the BsaDYs score with its biomarkers as stated above was trained to identify risk for acutely ill patients, based on the aDYs cohort with patients admitted to an emergency department with acute dyspnea.this probably also explains why only the BsMDc remained significantly related to outcomes when both scores (BsMDc and BsaDYs) were simultaneously included in the model.the BsMDc score is related to cci. the cci estimates survival in patients with multiple comorbidities. it consists of 17 items that are related to various morbidities that are associated with mortality.each morbidity has an associated weight (from 1 to 6), based on the adjusted risk of mortality and the sum of all the weights results in a single comorbidity score for a patient.the BsaDYs score on the contrary relates to 22 morbidities, where these morbidities are not adjusted or weighed for their severity.this is an important difference between the two scores, and perhaps might explain why morbidity scores adjusted to seriousness have a greater impact over time, than just a here and now presence of a previous or ongoing morbidity, when it comes to studies with a very long follow-up time.the BsMDc score also consists of 19 different biomarkers compared with 10 biomarkers for the BsaDYs score, which perhaps partly explains why BsMDc is so much stronger related to the outcomes.it is important to understand that inflammatory biomarkers are affected to varying degrees by acute illness from slight to very high increase (Zhang et al. 2017), where the values later can be lowered or normalized (chaikijurajai and tang 2020).On the contrary we believe that a low-grade increase in inflammatory biomarkers in a healthy population instead can be assumed to reflect genetic and more constant traits.thus, the two biomarker scores reflect and highlight different things, state versus trait, and thus can have their future role in completely different contexts.
We do however believe that it is not clinically practical to test and analyze such a large number of biomarkers as suggested in our results.a simple point-of-care test on whole blood with for example 3 different biomarkers is possible, with the aim of creating a test that can be clinically used to identify certain phenotypes in the population to predict future new onset of dyspneic diseases.in fact such testing already exist on the market with the option to customize 3 different biomarkers for a test analysis.
Our study might imply that in the future a screening consisting of a simple blood sample with analysis of some of biomarkers of the BsMDc in the middle-aged normal population could find people end phenotypes at risk for developing dyspnea morbidities and mortality in cardiovascular, respiratory and malignant diseases as well as-all-cause mortality.it can thus be used to robustly identify risk of future chF, cOPD and premature mortality in subjects free from cci and other major comorbidities at baseline. in its extension, it could thus prove to be an important tool within future preventive medicine.We believe that a selection of biomarkers from the BsMDc score primarily can be used for primary prevention, while a selection of biomarkers the BsaDYs score might have a role when it comes to secondary and tertiary prevention.Maybe it will be possible to reverse the course of  action through various early preventive measures aimed at preventing such serious dyspnea diseases as chF and cOPD before the diseases even have seriously started, as well as preventing deaths in cVD, cancer and respiratory diseases.in that case, a simple blood test could perhaps cost-effectively prevent a lot of diseases in society, as well as individual suffering.this remains to be shown in future research.

Strengths and limitations
it is a strength that we have a big cohort with a very long follow-up.the finding that a biological fingerprint of comorbidities (i.e.BsMDc) predicts morbidity and mortality in a population free from such comorbidities at baseline, shows that the BsMDc can identify clinically significant risks that are not possible to identify with anamnesis and standard clinical risk factors, which is a novel finding.
One limitation of this study is the low number of participants registered with a cci defined disease.however, despite this, the number of diseases seems sufficient to find associations of clinical importance with several biomarkers.We emphasize that even we found both strong and statistically independent associations between the BsMDc and all 6 outcomes, no conclusions regarding causality could be drawn owing to the observational nature of our study.Yet, from a clinical point of view when risk stratification of higher importance, the findings of the study are interesting and important despite that causality could not be claimed.
another major limitation is the lack of possible replication in an independent cohort, as the mortality risk has changed over time, especially for cVD mortality, as the MDc was created during the 1990-ies.however, in all studies with a long follow-up this factor is present.

Conclusion
a selection of biomarkers from the presented biomarker score of comorbidity burden could be useful in identifying people at risk for developing new-onset chF and cOPD, deaths in cVD, respiratory diseases, malignancies as well as all-cause deaths many years in advance in subjects free from comorbidities.Further studies are needed to extract some of the most significant biomarkers, and test specificity and sensitivity for the combination of these.Whether risks identified can be reverted with preventive actions remains to be elucidated.

Clinical significance
identifying a biomarker score of comorbidity burden is of importance to be able to identify people at risk for incident cardiovascular and respiratory diseases and mortality.

Figure 1 .
Figure 1.Flowchart of excluded and included patients.

Figure 2 .
Figure 2. Predictive ability of biomarker score (BsmDC) concerning future developing of CHF.

Figure 5 .
Figure 5. Predictive ability of biomarker score (BsmDC) score concerning death in cardiovascular diseases.

Table 2 .
Biomarkers included in the different biomarker scores.individualswithcompletebiomarker data for both the BsaDYs and the BsMDc score, which left us with 4005 individuals.Four percent of the individuals had one or more diseases as defined by cci. as we wanted to investigate the prognostic value of the biomarker comorbidity score in subjects without comorbidities, we excluded all individuals with presence of at least one cci comorbidity, as indicated from National swedish patient registrars.individuals with caD and diabetes mellitus (DM) that did not result in a cci index, were also excluded.For determination of prevalence of caD and DM also other registries than the national inpatient registry were used as described previously(schomburg et al.  2019).this finally left us with a cohort of n = 3616 individuals for further investigations and statistics.

Table 4 .
Cox regressions with BsaDYs and BsmDC-new with new-onset heart failure and CoPD as endpoints, n = 3616.