Refine
Has Fulltext
- yes (5)
Is part of the Bibliography
- yes (5)
Document Type
- Doctoral Thesis (5) (remove)
Language
- English (5) (remove)
Keywords
Institute
- Institut für Klinische Epidemiologie und Biometrie (5) (remove)
Sonstige beteiligte Institutionen
Introduction.
Mobile health (mHealth) integrates mobile devices into healthcare, enabling remote monitoring, data collection, and personalized interventions. Machine Learning (ML), a subfield of Artificial Intelligence (AI), can use mHealth data to confirm or extend domain knowledge by finding associations within the data, i.e., with the goal of improving healthcare decisions. In this work, two data collection techniques were used for mHealth data fed into ML systems: Mobile Crowdsensing (MCS), which is a collaborative data gathering approach, and Ecological Momentary Assessments (EMA), which capture real-time individual experiences within the individual’s common environments using questionnaires and sensors. We collected EMA and MCS data on tinnitus and COVID-19. About 15 % of the world’s population suffers from tinnitus.
Materials & Methods.
This thesis investigates the challenges of ML systems when using MCS and EMA data. It asks: How can ML confirm or broad domain knowledge? Domain knowledge refers to expertise and understanding in a specific field, gained through experience and education. Are ML systems always superior to simple heuristics and if yes, how can one reach explainable AI (XAI) in the presence of mHealth data? An XAI method enables a human to understand why a model makes certain predictions. Finally, which guidelines can be beneficial for the use of ML within the mHealth domain? In tinnitus research, ML discerns gender, temperature, and season-related variations among patients. In the realm of COVID-19, we collaboratively designed a COVID-19 check app for public education, incorporating EMA data to offer informative feedback on COVID-19-related matters. This thesis uses seven EMA datasets with more than 250,000 assessments. Our analyses revealed a set of challenges: App user over-representation, time gaps, identity ambiguity, and operating system specific rounding errors, among others. Our systematic review of 450 medical studies assessed prior utilization of XAI methods.
Results.
ML models predict gender and tinnitus perception, validating gender-linked tinnitus disparities. Using season and temperature to predict tinnitus shows the association of these variables with tinnitus. Multiple assessments of one app user can constitute a group. Neglecting these groups in data sets leads to model overfitting. In select instances, heuristics outperform ML models, highlighting the need for domain expert consultation to unveil hidden groups or find simple heuristics.
Conclusion.
This thesis suggests guidelines for mHealth related data analyses and improves estimates for ML performance. Close communication with medical domain experts to identify latent user subsets and incremental benefits of ML is essential.
1 Summary
Left ventricular (LV) ejection fraction (EF) and global longitudinal strain (GLS) are the most commonly used measures of LV function. Yet, they are highly dependent on loading conditions since higher afterload yields lower systolic deformation and thereby a lower LVEF and GLS – despite presumably unchanged LV myocardial contractile strength. Invasive pressure-volume loop measurements represent the reference standard to assess LV function, also considering loading conditions. However, this procedure cannot be used in serial investigations or large sample populations due to its invasive nature. The novel concept of echocardiography-derived assessment of myocardial work (MyW) is based on LV pressure-strain loops, may be a valuable alternative to overcome these challenges, and may also be used with relative ease in large populations. As MyW also accounts for afterload, it is considered less load-dependent than LVEF and GLS.
The current PhD work addresses the application and clinical characterization of MyW, an innovative echocardiographic tool. As the method is new, we focused on four main topics:
(a) To establish reference values for MyW indices, i.e., Global Work Index (GWI), Global Constructive Work (GCW), Global Wasted Work (GWW), and Global Work Efficiency (GWE); we addressed a wide age range and evaluated the association of MyW indices with age, sex and other clinical and echocardiography parameters in apparently cardiovascular healthy individuals.
(b) To investigate the impact of cardiovascular (CV) risk factors on MyW indices and characterize the severity of subclinical LV deterioration in the general population.
(c) To assess the association of the LV geometry, i.e., LV mass and dimensions, with MyW indices.
(d) To evaluate in-hospital dynamics of MyW indices in patients hospitalized for acute heart failure (AHF).
For the PhD thesis, we could make use of two larger cohorts:
The STAAB population-based cohort study prospectively recruited and phenotyped a representative sample (5,000 individuals) of the general population of the City of Würzburg, aged 30-79 years and free from symptomatic heart failure at the time of inclusion. We focused on the first half of the study sample (n=2473 individuals), which fulfilled the anticipated strata regarding age and sex.
The Acute Heart Failure (AHF) Registry is a prospective clinical registry recruiting and phenotyping consecutive patients admitted for decompensated AHF to the Department of Medicine I, University Hospital Würzburg, and observing the natural course of the disease. The AHF Registry focuses on the pathophysiological understanding, particularly in relation to the early phase after cardiac decompensation, with the aim to improve diagnosis and better-tailored treatment of patients with AHF. For the current study, we concentrated on patients who provided pairs of echocardiograms acquired early after index hospital admission and prior to discharge.
The main findings of the PhD thesis were:
From the STAAB cohort study, we determined the feasibility of large-scale MyW derivation and the accuracy of the method. We established reference values for MyW indices based on 779 analyzable, apparently healthy participants (mean age 49 ± 10 years, 59% women), who were in sinus rhythm, free from CV risk factors or CV disease, and had no significant LV valve disease. Apart from GWI, there were no associations of other MyW indices with sex. Further, we found a disparate association with age, where MyW showed stable values until the age of 45 years, with an upward shift occurring beyond the age of 45. A higher age decade was associated with higher GWW and lower GWE, respectively. MyW indices only correlated weakly with common echocardiographic parameters, suggesting that MyW may add incremental information to clinically established parameters.
Further analyses from the STAAB cohort study contributed to a better understanding of the impact of CV risk factors on MyW indices and the association of LV geometry with LV performance. We demonstrated that CV risk factors impacted selectively on GCW and GWW. Hypertension appears to profoundly compromise the work of the myocardium, in particular, by increasing both GCW and GWW. The LV in hypertension seems to operate at a higher energy level yet lower efficiency. Other classical CV risk factors (Diabetes mellitus, Obesity, Dyslipidemia, Smoking) – independent of blood pressure – impacted consistently and adversely on GCW but did not affect GWW. Further, all CV risk factors affected GWE adversely.
We observed that any deviation from a normal LV geometric profile was associated with alterations on MyW. Of note, MyW was sensitive to early changes in LV mass and dimensions. Individuals with normal LV geometry yet established arterial hypertension exhibited a MyW pattern that is typically found in LV hypertrophy. Therefore, such a pattern might serve as an early sign of myocardial damage in hypertensive heart disease and might aid in risk stratification and primary prevention.
From the AHF Registry, we selected individuals with serial in-hospital echocardiograms and described in-hospital changes in myocardial performance during recompensation. In patients presenting with a reduced ejection fraction (HFrEF), decreasing N-terminal pro-natriuretic peptide (NT-proBNP) levels as a surrogate of successful recompensation were associated with an improvement in GCW and GWI and consecutively in GWE. In contrast, in patients presenting with a preserved ejection fraction (HFpEF), there was no significant change in GCW and GWI. However, unsuccessful recompensation, i.e., no change or an increase in NT-proBNP levels, was associated with an increase in GWW. This suggests a differential myocardial response to de- and recompensation depending on the HF phenotype.
Further, GWW as a surrogate of inappropriate LV energy consumption was elevated in all patients with AHF (compared to reference values) and was not associated with conventional markers as LVEF or NT-proBNP. In an exploratory analysis, GWW predicted the risk of death or rehospitalization within six months after discharge. Hence, GWW might carry incremental information beyond conventional markers of HF severity.
Background
Tobacco smoking is accountable for more than one in ten deaths in patients with cardiovascular disease. Thus, smoking cessation has a high priority in secondary prevention of coronary heart disease (CHD). The present study meant to assess smoking cessation patterns, identify parameters associated with smoking cessation and investigate personal reasons to change or maintain smoking habits in patients with established CHD.
Methods
Quality of CHD care was surveyed in 24 European countries in 2012/13 by the fourth European Survey of Cardiovascular Disease Prevention and Diabetes. Patients 18 to 79 years of age at the date of the CHD index event hospitalized due to first or recurrent diagnosis of coronary artery bypass graft, percutaneous coronary intervention, acute myocardial infarction or acute myocardial ischemia without infarction (troponin negative) were included. Smoking status and clinical parameters were iteratively obtained a) at the cardiovascular disease index event by medical record abstraction, b) during a face-to-face interview 6 to 36 months after the index event (i.e. baseline visit) and c) by telephone-based follow-up interview two years after the baseline visit. Parameters associated with smoking status at the time of follow-up interview were identified by logistic regression analysis. Personal reasons to change or maintain smoking habits were assessed in a qualitative interview and analyzed by qualitative content analysis.
Results
One hundred and four of 469 (22.2%) participants had been classified current smokers at the index event and were available for follow-up interview. After a median observation period of 3.5 years (quartiles 3.0, 4.1), 65 of 104 participants (62.5%) were classified quitters at the time of follow-up interview. There was a tendency of diabetes being more prevalent in quitters vs non-quitters (37.5% vs 20.5%, p=0.07). Higher education level (15.4% vs 33.3%, p=0.03) and depressed mood (17.2% vs 35.9%, p=0.03) were less frequent in quitters vs non-quitters. Quitters more frequently participated in cardiac rehabilitation programs (83.1% vs 48.7%, p<0.001). Cardiac rehabilitation appeared as factor associated with smoking cessation in multivariable logistic regression analysis (OR 5.19, 95%CI 1.87 to 14.46, p=0.002). Persistent smokers at telephone-based follow-up interview reported on addiction as wells as relaxation and pleasure as reasons to continue their habit. Those current and former smokers who relapsed at least once after a quitting attempt, stated future health hazards as their main reason to undertake quitting attempts. Prevalent factors leading to relapse were influence by their social network and stress. Successful quitters at follow-up interview referred to smoking-related harm done to their health having had been their major reason to quit.
Interpretation
Participating in a cardiac rehabilitation program was strongly associated with smoking cessation after a cardiovascular disease index event. Smoking cessation counseling and relapse prophylaxis may include alternatives for the pleasant aspects of smoking and incorporate effective strategies to resist relapse.
In several countries, a decline in mortality, case-fatality and recurrence rates of stroke was observed. However, studies investigating sex-specific and subtype-specific (pathological and etiological) time trends in stroke mortality, case-fatality and recurrence rates are scarce, especially in Germany. The decline in ischemic stroke mortality and case-fatality might be associated with the high quality of acute care of ischemic stroke, but the exact determinants of early outcome remains unknown for Germany.
Therefore, as first step of this thesis, we investigated the time trends of subtype- and sex-specific age- standardized stroke mortality rates in Germany from 1998 to 2015, by applying joinpoint regression on official causes of death statistics, provided by the Federal Statistical Office. Furthermore, a regional comparison of the time trends in stroke mortality between East and West was conducted. In the second step, time trends in case-fatality and stroke recurrence rates were analyzed using data from a population- based stroke register in Germany between 1996 and 2015. The analysis was stratified by sex and etiological subtype of ischemic stroke. In the third step, quality of stroke care and the association between adherence to measures of quality of acute ischemic stroke care and in-hospital mortality was estimated based on data from nine regional hospital-based stroke registers in Germany from the years 2015 and 2016.
We showed that in Germany, age-standardized stroke mortality declined by over 50% from 1998 to 2015 both, in women and men. Stratified by the pathological subtypes of stroke, the decrease in mortality was larger in ischemic stroke compared to hemorrhagic stroke. Different patterns in the time trends of stroke were observed for stroke subtypes, regions in Germany (former Eastern part of Germany (EG), former Western part of Germany (WG)) and sex, but in all strata a decline was found. By applying joinpoint regression, the number of changes in time trend differed between the regions and up to three changes in the trend in ischemic stroke mortality were detected. Trends in hemorrhagic stroke were in parallel between the regions with up to one change (in women) in joinpoint regression. Comparing the regions, stroke mortality was higher in EG compared to WG throughout the whole observed time period, however the differences between the regions started to diminish from 2007 onwards.
Further it was found that, based on the population-based Erlangen Stroke Project (ESPro), case-fatality and recurrence rates in ischemic stroke patients are still high in Germany. 46% died and 20% got a recurrent stroke within the first five years after stroke. Case-fatality rates declined statistically significant from 1996 to 2015 across all ischemic stroke patients and all etiological subtypes of ischemic stroke. Based on Cox regression no statistically significant decrease in stroke recurrence was observed.
Based on the pooled data of nine regional hospital-based stroke registers from the years 2015 and 2016 covering about 80% of all hospitalized stroke patients in Germany, a high quality of care of acute ischemic stroke patients, measured via 11 evidence-based quality indicators (QI) of process of care, was observed. Across all registers, most QI reached the predefined target values for good quality of stroke care. 9 out of 11 QI showed a significant association with 7-day in-hospital mortality. An inverse linear association between overall adherence to QI and 7-day in-hospital mortality was observed.
In conclusion, stroke mortality and case-fatality showed a favorable development over time in Germany, which might partly be due to improvements in acute treatment. This is supported by the association between overall adherence to quality of care and in-hospital mortality. However, there might be room for improvements in long-term secondary prevention, as no clear reduction in recurrence rates was observed.
Chronic Kidney Disease as an Important Co-morbid Condition in Coronary Heart Disease Patients
(2019)
In patients with coronary heart disease (CHD) the control of the modifiable “traditional” cardiovascular risk factors such as hypertension, dyslipidemia, diabetes, achieving/maintaining normal body weight and smoking cessation is of major importance to improve prognosis. Guideline recommendations for secondary CHD prevention include specific treatment targets for blood pressure, lipid levels, and markers of glucose metabolism for both younger and older patients. Chronic kidney disease (CKD) has been identified as a “non-traditional” risk factor for worse outcome in CHD patients, as it is associated with a markedly increased risk for subsequent CV events and mortality.
The specific objectives of the current thesis-project are to investigate (a) the quality of care in a recent sample of German CHD patients and to investigate variation of risk factor control between younger and elder patients (≤70 versus >70 years), (b) to analyze the prevalence of CKD across Europe in stable CHD patients in the outpatient setting and during a hospital stay for CHD, (c) to investigate the level of awareness of CKD in German CHD patients and their treating physicians.
Data from the European-wide EUROASPIRE IV study were used that include data on 7998 CHD patients in the ambulatory setting (study visit) and during a hospital stay for CHD (index). The German EUROASPIRE IV study center in Würzburg recruited 536 patients in 2012-2013. Risk factor control was compared against the current recommendations of the European Society of Cardiology. CKD was described by stages of glomerular filtration rate (eGFR) and albuminuria. German patients were asked in an additional kidney specific module whether they have ever been told by a physician about renal impairment. The fact that CKD or acute kidney injury (AKI) was mentioned in prominent parts of the hospital discharge letter as well as correct ICD-coding of CKD or AKI served as a proxy for physician’s awareness of CKD.
The majority of German CHD patients was treated with the recommended drug therapies including e.g. β-blockers, anti-platelets and statins. However, treatment targets for blood pressure and LDL-cholesterol levels were not achieved in many patients (45% and 53%, respectively) and glycemic control in diabetic CHD patients with HbA1-levels <7% was insufficient (61%). A minority of patients reported on current smoking (10%), but unhealthy life-styles e.g. overweight/obesity (85%/37%) were frequent. Patterns of care differed between younger and older CHD patients while older patients were less likely to receive the recommended medical CHD-therapy, were more likely to have uncontrolled blood pressure and also to be diabetic. However, a greater proportion of diabetic patients >70 years was achieving the HbA1c target, and less elder patients were current smokers or were obese. About 17% of patients on average had CKD (eGFR< 60 ml/min/1.73m²) in the entire European sample at the study visit, and an additional 10% had albuminuria despite preserved eGFR, with considerable variation among countries. Impaired kidney function was observed in every fifth patient admitted for CHD in the entire European dataset of the EUROASPIRE IV study. Of the German CHD patients with CKD at the study visit, only a third were aware of their renal impairment. A minority of these patients was being seen by nephrologists, however, with a higher likelihood of CKD awareness and specialist care in more advanced stages of CKD. About a third of patients admitted for CHD showed either CKD or AKI during the hospital stay, but the discharge letter mentioned chronic or acute kidney disease only in every fifth of these patients. In contrast, correct ICD coding of CKD or AKI was more complete, but still suboptimal.
In summary, quality of secondary prevention in German CHD patients indicates considerably room for improvement, with life-style modifications may become an even greater factor in prevention campaigns than medical treatment into certain target ranges. Preventive therapies should also consider different needs in older individuals acknowledging physical and mental potential, other comorbidities and drug-interactions with co-medication. CKD is common in CHD patients, not only in the elderly. Since CHD and CKD affect each other and impact on worse prognosis of each other, raising the awareness of CKD among patients and physicians and considering CKD in medical therapy may improve prognosis and slow disease progression of CHD as well as CKD.