Introduction
Heart failure (HF) affects approximately 1–2% of the adult population in developed countries and has a prevalence of over 10% among people aged over 70 years.1 The prognosis for the condition is currently poor. While acute HF is a sudden onset of symptoms, chronic HF is characterized by repeated hospitalizations and has a high 30 day re-hospitalization rate2 as well as high 1 year mortality3 ranging from 15% to 30% and a 5 year rate of up to 75% in specific populations.4
Healthcare trajectories are essential monitoring tools, particularly to monitor care for chronic diseases such as diabetes, organ failure, coronary artery disease, and/or acute coronary syndrome (ACS). Models of healthcare trajectories have been developed and are key to reducing readmission rates and improving the quality of both ambulatory and inpatient care.5 They help construct a rule-based prediction model that provides data concerning trends seen in patient pathways between different medical units.6 For instance, in Pinaire et al. (2017), a predictive model of ACS trajectories was used to simulate future disease progression in order to anticipate subsequent healthcare needs.7
Claim databases are valuable data sources for healthcare trajectory analyses because of their longitudinal properties. Standard statistical analyses cannot be used in this context because the assumptions of the models are not verified. Developing new methods has to be encouraged to take into account specific characteristics of such data, including the importance of the order in which events occur over time and the time between two events.
Methodological research has been carried out on these bases to develop sequence similarity measures for clustering.8,9 These sequence clustering techniques have already been used to identify typical healthcare trajectories after bariatric surgery10 or typical treatment sequences in ambulatory care for patients with HF.11
Predicting the mortality rate of HF is necessary for clinicians to make optimal decisions during the therapeutic process. A recent observational study evaluated healthcare utilization for people identified with HF and showed the complexity of patient pathways.12 However, to our knowledge, no study has yet evaluated different healthcare trajectories for patients with HF, nor sequence similarity measures and their association with mortality.
Healthcare claims data from the national medical insurance schemes can be used to classify healthcare trajectories.10 In France, the national healthcare claims database known as the Programme de Médicalisation des Systèmes d'Information (PMSI) includes all reimbursed hospitalizations performed in both public and private hospitals. Diagnosis-related group (DRG) sequences are available by amalgamating hospital stays in order to identify the chronological pattern of hospitalizations. The primary objectives of this study were to analyse the nationwide healthcare trajectories of patients with HF in France, 2 years after their first hospitalization, and to measure sequence similarities. Secondary objectives were to identify the association between trajectories and the risk of mortality [by extending an analysis process proposed by Pinaire et al. (2017)7 to take into account the order of re-hospitalizations and time to death].
Methods
Study design
A national retrospective observational study was conducted using data extracted from the Echantillon Généraliste des Bénéficiaires database between 1 January 2008 and 31 December 2018. The design of the study is illustrated in Supporting Information, Figure S1. Along with the PMSI, these databases are organized into a comprehensive digital data source compiling the total consumption of outpatient and inpatient care in both public and privately managed facilities. Insured patients are assigned a unique identifier that allows their healthcare utilization to be tracked throughout their lifetime. This study also used hospital data, which were exhaustive and collected as standardized discharge reports.
Each hospital stay entered in the database was categorized into the following fields: (i) anonymized patient identification; (ii) entry and exit date; (iii) length of stay (LoS); (iv) primary and associated diagnoses based on the International Classification of Disease (10th edition) (ICD-10); and (v) the therapeutic procedures received. All the information was incorporated into the DRG. Patient death and date of death included both in-hospital and out-of-hospital occurrences.
Inclusion/exclusion criteria
Eligible patients were over 18 years old with an HF hospitalization identified from 1 January 2010 to 31 December 2016. HF hospitalization was defined as a stay itemized under a specific ICD-10 code of HF or a DRG code (see the selection criteria in Supporting Information, Table S1). The date of the first hospitalization during the study period was used as the index date. Patients were followed up from the index date to either the patient's death or the end of the study. Eligible patients were screened for any previous HF hospitalization occurring between 1 January 2008 and 31 December 2009, and if an occurrence was found, they were excluded.
Statistical analyses
Quantitative variables were described using mean and standard deviation or median and interquartile range. Qualitative variables were described using numbers and percentages. Overall survival (OS) rates were analysed using Kaplan–Meier survival curves and compared using the log-rank test. The steps of the statistical analysis are explained below and shown in Table 1. The statistical methodology used in this study was an exploratory pathway and non-inferential analysis involving the identification of healthcare trajectories possibly associated with mortality. Therefore, no assumptions were made a priori. The trajectories were compared with each other in relative terms.
Table 1 Study design and steps of statistical analyses
| Study design | |
| Step of statistical analyses | Population |
| Step 1: Survival and re-hospitalization related to cardiovascular diseases (CVDs) analysis |
All patients Patients with a first hospitalization for heart failure (HF) between 2010 and 2016 |
| Step 2: Construction of hospitalization sequencesa |
Only re-hospitalized patients Patients with a first hospitalization for HF between 2010 and 2016 AND With at least one CVD related re-hospitalization |
| Step 3: Identification of frequent hospitalization sequences | |
| Step 4: Calculation of similarity score between patient hospitalization sequences and each frequent hospitalization sequence in order to evaluate the prognosis value of frequent sequences | |
| Step 5: Analysis of the association between frequent hospitalization sequences and death and identification of the 20 most frequent sequences associated with death and survival |
Hospitalization sequence analyses (Steps 2–5)
The primary outcome was to identify the frequent healthcare trajectories defined by recurrent hospitalizations. Hospitalization trajectory data were firstly identified for each patient, and then, in order to identify healthcare trajectories predictive of mortality, a similarity score was calculated between each patient's hospitalization trajectory and frequency so as to quantify the similarity of sequences. The scores obtained for each sequence were used as covariates in a multivariate model to predict mortality.7 We hypothesized that frequent courses are good predictors of mortality, taking into account patient age and gender.
Construction of hospitalization sequences (Step 2)
One hospitalization sequence was defined for each patient from the index date to either patient death or the end of the study. These sequences were identified by the five first digits of each patient's DRG code and were then sorted sequentially according to date. We focused on hospitalizations associated with cardiac diseases only. The list of selected DRGs is presented in Supporting Information, Table S2.
Identification of frequent sequences (Step 3)
Sequences were identified from hospitalization data using a pattern mining algorithm that identifies recurrent character strings based on a fixed minimum threshold. As we had long and dense sequences, a Co-occurance Map sequentiel Pattern Mining (CM-SPAM) algorithm was used to identify frequent sequences. This algorithm works by vertically extracting patterns, making data analysis faster and less expensive. CM-SPAM was carried out using SPMF (V2.42) with the support of 1%.13
Scoring/similarities (Step 4)
For each patient, a score was calculated to quantify the similarity between their hospitalization sequence and each of the frequency sequences using the Smith–Waterson algorithm. This algorithm allowed us to directly compare the successions of DRGs by considering their order of appearance in the sequence. The scores obtained were between 0 and 1. The scores were equal to 0 if none of the DRGs of the hospitalization sequence were present in the studied frequency sequence and to 1 if the two sequences were identical. The text alignment library (V0.1.2) of R (V4.0.3) was used.
Prediction model (Step 5)
A gradient boosting algorithm for survival analysis with Cox's partial likelihood as the loss function was used to predict OS and identify frequent sequences with lower or high mortality. The event of interest was the delay between the first HF hospitalization and death/end of study. At the first HF hospitalization, age, gender, and the similarity score of each frequency sequence were used as input features. This algorithm implements gradient boosting with a regression tree-based learner. It follows the strength-in-numbers principle by combining the predictions of multiple base learners to obtain a comprehensive overall model. The predictions were combined in a manner in which the addition of each base model improved the overall model. This algorithm intrinsically contains a random part to avoid over-fitting and to ensure that we had robust results; it was run 30 times. Sensitivity analyses were carried out using other methods of machine learning for OS, such as random forest or CoxBoost.
For each repetition, feature importance was calculated. The weighted average of the 20 feature importance calculations was associated with each input feature, given that the greater the weight, the greater the impact of the variable in predicting survival. As these weights could be interpreted only in relative terms within the same analysis, they are not presented in our findings and are detailed in the supporting information. To evaluate the direction of the covariate effect, partial dependency graphs were used. Only the results of input features with the same direction of effect in the 30 repetitions were interpreted. Results were ordered according to the average weights from the feature importance calculations. These analyses were performed using Python (V3.8.6) with scikit-survival, ELI5, and PDP modules installed.
Results
Population characteristics
Between 2010 and 2016, 12 026 patients were identified with a first HF hospitalization; however, 480 were ineligible due to no healthcare consumption or unusable data. Fifty-eight patients with a single hospital admission for Code 05M22 (other CMD05 conditions with death: stays of <2 days) were also excluded. Therefore, 11 488 patients were included in this study (Supporting Information, Figure S2). The mean age of patients was 78 ± 13 years, and 49.5% were male (Table 2). The follow-up period averaged 2.9 years, and an OS analysis after the first HF hospitalization showed a mortality rate of 31.7% in the first year, 41.5% at 2 years, 57.2% at 3 years, and 78.8% at 5 years (Figure 1).
Table 2 Patient characteristics with re-hospitalizations
| Patient characteristics | Included patients |
| Gender | |
| Male | 5685 (49.5) |
| Female | 5803 (50.5) |
| Age at the first HF hospitalization, mean ± SD | 78.0 ± 13.2 |
| Follow-up duration (years), mean ± SD | 2.9 ± 1.3 |
| Re-hospitalizations overall follow-up | |
| At least one hospitalization during the overall follow-up after the first HF hospitalization | 9220 (80.3) |
| At least one hospitalization related to CVDs* during the overall follow-up after the first HF hospitalization | 5704 (49.7) |
| Re-hospitalizations during the first year following the first HF hospitalization | |
| At least one hospitalization during the first year following the first HF hospitalization | 7134 (62.1) |
| At least one hospitalization related to CVDs during the first year following the first HF hospitalization | 3837 (33.4) |
| At least one hospitalization related to HF during the first year following the first HF hospitalization | 1436 (12.5) |
[IMAGE OMITTED. SEE PDF]
During follow-up, 80% of patients were readmitted at least once for any reason, and almost half (49.7%) were re-hospitalized for concerns related to cardiovascular diseases (CVDs). Re-hospitalization analyses therefore covered 5704 patients. During the first year following the first HF re-hospitalization, 62.1% of patients had at least one re-hospitalization (for any reason), 33.4% for CVDs, and 12.5% for HF specifically (Table 2). On average, patients were hospitalized 1.32 times and 2.68 times for patients re-hospitalized within 2 years of the first hospitalization (Figure 1). Among the patients who died, most had a previous hospitalization (Code 05M, which is a medical hospitalization for circulatory system disease); however, the reverse was not true. After a re-hospitalization for Code 05M, 27.4% of patients died, whereas 46.6% of patients re-hospitalized for Code 04M (which is for pulmonary oedema and respiratory distress) died (Figure 2). Hospitalization for Code 04M seemed the riskiest.
[IMAGE OMITTED. SEE PDF]
Healthcare trajectories and similarities
In regard to healthcare trajectories, the analysis of hospitalization sequences was based on 5704 patients. In total, 2222 distinct sequences were identified. From these sequences, the CM-SPAM algorithm selected 89 recurrent sequences. Similarity scores between the sequence of each patient and the most frequent sequences were calculated. These similarity scores are presented for the 20 most frequent sequences (Figure 3).
[IMAGE OMITTED. SEE PDF]
After running the gradient boosting algorithm 30 times, 1707 sequences were identified, of which 21 sequences were associated with a good prognosis (see Supporting Information, Table S3) and 15 sequences with a bad prognosis (Supporting Information, Table S4). Fifty-three frequent sequences lacked a clear sense of association (n = 43) or had contradictory associations between repetitions (n = 10) and were not interpreted (Supporting Information, Table S5). In all models, age and gender were the key predictors of mortality (permutation feature importance: 0.099 ± 0.00078 and 0.0087 ± 0.00018, respectively; weights could be interpreted in relative terms). Age at first HF hospitalization was positively associated with mortality, and males had a poorer prognosis. We identified 21 sequences associated with a good prognosis and 15 with a poor prognosis. Therapeutic interventions were the elements most often found in sequences associated with a good prognosis. The presence of re-hospitalization for HF within a sequence was the main element associated with a poor prognosis.
Sequences associated with a good prognosis
The most significant sequences associated with a good prognosis are shown in Figure 4. These trajectories involved patients who had received surgical and non-surgical device treatments such as permanent pacemaker placements (without acute myocardial infarction, congestive HF, or shock), permanent pacemaker replacements, placement of a cardiac defibrillator, vascular stents (without myocardial infarction) followed by vascular diagnostic procedures, vascular diagnostic procedures and vascular stents (without myocardial infarction), cardiac valve bio-prosthetic installation by vascular route, major treatments for arrhythmias by vascular route and other treatments for arrhythmias by vascular route, valve replacement surgery with extracorporeal circulation (without cardiac catheterization or coronary angiography), and lastly, coronary artery bypass grafts (without cardiac catheterization or coronary angiography).
[IMAGE OMITTED. SEE PDF]
Other patterns of hospitalization were associated with a good prognosis, namely, (i) chest pain, (ii) hypertension, and (iii) syncope/fainting. All sequences associated with a good prognosis are shown in Supporting Information, Table S3.
Sequences associated with a bad prognosis
The most significant sequences associated with a bad prognosis for HF are shown in Figure 4. Healthcare trajectories included (i) hospitalization for HF before and/or after other reasons for hospitalizations; (ii) vascular diagnostic procedures followed by HF and circulatory shock; (iii) chest pain followed by HF and circulatory shock; (iv) coronary atherosclerosis followed by HF and circulatory shock; (v) HF and circulatory shock followed by placement of a cardiac defibrillator; (vi) placement of a cardiac defibrillator followed by HF and circulatory shock; and (vii) HF and circulatory shock followed by vascular diagnostic procedures. All sequences associated with a bad prognosis are shown in Supporting Information, Table S4.
Discussion
In this study, we showed that a very poor prognosis continued to exist in the HF population, with a 1 year mortality rate >30%. In all our models, age was positively associated with mortality, and gender (male) was associated with poorer prognoses. Healthcare trajectories, including non-surgical device treatments, valve replacements, and atrial fibrillation ablation, were associated with a better prognosis, except in cases where these invasive treatments preceded or followed hospitalization for cardiac decompensation. The predominant negative prognosis sequences were mostly those that included HF-related hospitalizations before or after other-related hospitalizations. We also highlighted the heterogeneity and complexity of healthcare trajectories after a patient's first hospitalization for HF. Re-hospitalization for HF has a predominant place in sequences associated with higher mortality.
Our study population was comparable with the literature, having older patients, more females than males, and women older than men,3 with poor prognoses and a high rate of hospitalization and mortality.
The mortality rate in our cohort was high, with 31.6% at 1 year, 41.5% at 2 years, and 78.7% at 5 years. These rates are similar to those of other studies, such as Shah et al. (2017), which showed 75.4% at 5 years, covering North America.14 This rate was also comparable with the UK retrospective population-based study, in which the mortality rate after a first diagnosis of HF was 32% at 1 year.15 Our findings were higher than those shown in a Spanish study (11.3% at 1 year).16 One explanation for this could be the inclusion criteria of our study, which was an HF-related hospitalization, an event with an inherently bad prognostic value. The high rate of re-hospitalization observed (33.5% for all CVDs in the first year) was comparable with Groenewegen et al. (2020).17 Higher rates were reported in Constantinou et al. (2021) using administrative data. In this study, Constantinou et al. (2021) found that 31.8% were readmitted at least once for HF over the 1 year follow-up2; however, patients may have been managed for HF before. Our results were consistent with the European Society of Cardiology (ESC) guidelines reporting that post-discharge 1 year mortality can be 25–30%, with up to more than 45% deaths or readmission rates.1
Our findings on gender, with women having fewer hospital readmissions for CVDs, were also similar to those of other studies such as Nozaki et al. (2017).18 The two parameters most strongly associated with mortality in our study were advanced age and gender being male (independent of age), which were similarly reported in Simpson et al. (2020).19
When interpreting the prognosis of patients included in this study, it should be noted that these were patients diagnosed with HF only during hospitalization, a factor with a poor prognosis, and that it was the first episode of HF, excluding previous HF events with a worse prognosis.
To date, the prognosis of HF has been estimated by the presence of previous hospitalizations for HF. Solomon et al. (2007) reported mortality rates to be inversely proportional to the time since the last hospitalization.20 In symptomatic chronic patients, HF hospitalization frequency has been shown to be a key predictor of death after discharge.20 In our study, we did not observe this particularly, and the risk of mortality was high in the immediate post-discharge period, which corresponded to the findings in Solomon et al. (2007).20
Among the sequences associated with a good prognosis, hospitalizations for therapeutic interventions had the greatest weight, which supports recommendations. Healthcare trajectories and the inclusion of non-surgical device treatments, valve replacements, and atrial fibrillation ablation were associated with a better prognosis, except in cases where these invasive treatments preceded or followed hospitalization for cardiac decompensation. Regarding implantable cardioverter-defibrillators (ICDs), our results matched two previous studies. The first focused on hospitalizations occurring within 30 days before an ICD therapy replacement and its association with death,21 and the second reported increasing survival rates in patients at risk of sudden cardiac death due to ventricular tachyarrhythmia ICD therapy.1 A recently published meta-analysis pooling data from all randomized control trials and tested primary-prevention ICD over the past two decades (including the DANISH trial) also confirmed a significant reduction of all-cause mortality associated with ICD use in patients with non-ischaemic cardiomyopathy.22 Nevzorov et al. (2018) evaluated the factors associated with mortality after ICD implantation23; however, prior hospitalizations for HF were not included in the study variables. If the benefits of the non-surgical device treatments are well in line with the latest ESC guidelines, then our main findings could show the key prognostic value of HF re-hospitalization and the decisive prognostic value compared with other sequences.
The significance of the identified prognostic factors in this study was that when including HF readmissions, the prognostic value of a trajectory was oriented towards a worsened prognosis (12 of the 15 pathways associated with a poor prognosis included HF readmission). This was similarly found in an observational study from the United States that demonstrated that longer HF readmission-free periods were associated with decreased risks of in-hospital mortality.24 Similarly, our results support the fact that the number of HF-related hospitalizations is a strong predictor of mortality in HF patients, even though Setoguchi et al. (2007) showed that survival was inversely correlated with each HF hospitalization episode.25
There were both strengths and limitations to this study. Regarding strengths, our cohort size was large, and the observational period over 2 years was longer compared with previous literature. Regarding limitations, as our study was exploratory and descriptive, common administrative data coding limitations and missing clinical data to distinguish the type of HF were imposed. As this study was also retrospective in design, the findings may not draw causal relationships.
Second, patients were included based on their primary ICD-10 diagnosis or associated code of HF. We used this codification to characterize healthcare trajectories in order to have a unique code per hospitalization while also taking into account the primary diagnosis, any associated diagnoses, and the procedures performed. We did not look at hospital admission before HF diagnosis, and we did not compare healthcare trajectories between regions in France as other studies have performed, such as Rao et al. (2019).26 Third, we did not include data related to the LoS of readmitted patients as previously reported by Altibi et al. (2021)24 and focused only on sequences that may have incurred selection bias.
Our study may show that similarity measures of hospitalization sequences can be used to identify specific sequences associated with mortality using real-world data. The analysis process used applied methods from various fields of data analysis, including sequence pattern mining and machine learning. It may be worth noting that all patients were included in the analysis, such as those who died within 2 years of their first HF identification, for the sake of completeness and in order to highlight specific subgroups. This inclusion may have generated lead-time bias and immortal time bias, and a sensitivity analysis should be carried out in the future to exclude early deaths.
Conclusions
In conclusion, we indicate the necessary prognostic value of HF-related re-hospitalizations. A methodology specific to healthcare data was developed by extracting frequent trajectories and measuring their similarity for use in a survival machine learning analysis. We highlight the value of healthcare trajectories for frequent hospitalization sequences, mortality, and prognosis. Our work may be an essential tool for better identification of at-risk patients in order to increase and improve personalized care in the future.
Acknowledgements
The authors would like to thank Sarina Yaghobian and Marty Brucato from AcaciaTools for their proofreading and reviewing services.
Conflict of interest
The authors have no conflicts of interest related to this study.
Funding
This work was supported by the National Association for Research and Technology [Association Nationale de la Recherche et de la Technologie (ANRT)] in France (grant 2019/0065).
McDonagh TA, Metra M, Adamo M, Gardner RS, Baumbach A, Böhm M, et al. 2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J 2021;42:3599–3726. doi:
Constantinou P, Pelletier‐Fleury N, Olié V, Gastaldi‐Ménager C, JuillÈre Y, Tuppin P. Patient stratification for risk of readmission due to heart failure by using nationwide administrative data. J Card Fail 2021;27:266–276. doi:
Farré N, Vela E, Clèries M, Bustins M, Cainzos‐Achirica M, Enjuanes C, et al. Real world heart failure epidemiology and outcome: A population‐based analysis of 88,195 patients. Lazzeri C, ed. PLoS ONE 2017;12: [eLocator: e0172745]. doi:
Savarese G, Becher PM, Lund LH, Seferovic P, Rosano GMC, Coats AJS. Global burden of heart failure: A comprehensive and updated review of epidemiology. Cardiovasc Res 2023;118:3272–3287. doi:
Pinaire J, Azé J, Bringay S, Landais P. Patient healthcare trajectory. An essential monitoring tool: A systematic review. Health Inf Sci Syst 2017;5:1. doi:
Voors AA, Ouwerkerk W, Zannad F, van Veldhuisen DJ, Samani NJ, Ponikowski P, et al. Development and validation of multivariable models to predict mortality and hospitalization in patients with heart failure. Eur J Heart Fail 2017;19:627–634. doi:
Pinaire J, Azé J, Bringay S, Landais P. Infarctus du myocarde: quelles sont les trajectoires de soins pronostiques du décès à l'hôpital? In: Roussey C, ed. IC: Ingénierie des Connaissances. Caen, France; 2017:14–25.
Mannino M, Fredrickson J, Banaei‐Kashani F, Linck I, Alqurashi R. Development and evaluation of a similarity measure for medical event sequences. ACM Trans Manag Inf Syst 2017;8:1–26. doi:
Rivault Y, Le Meur N, Dameron O. A similarity measure based on care trajectories as sequences of sets. In: ten Teije A, Popow C, Holmes JH, Sacchi L, eds. Artificial Intelligence in Medicine. Cham: Springer International Publishing; 2017:278–282.
Charles‐Nelson A, Lazzati A, Katsahian S. Analysis of trajectories of care after bariatric surgery using data mining method and health administrative information systems. Obes Surg 2020;30:2206–2216. doi:
Vogt V, Scholz SM, Sundmacher L. Applying sequence clustering techniques to explore practice‐based ambulatory care pathways in insurance claims data. Eur J Public Health 2018;28:214–219. doi:
Feldman SF, Lesuffleur T, Olié V, Gastaldi‐Ménager C, Juillière Y, Tuppin P. French annual national observational study of 2015 outpatient and inpatient healthcare utilization by approximately half a million patients with previous heart failure diagnosis. Arch Cardiovasc Dis 2021;114:17–32. doi:
Fournier‐Viger P, Gomariz A, Campos M, Thomas R. Fast vertical mining of sequential patterns using co‐occurrence information. In: Tseng VS, Ho TB, Zhou Z‐H, Chen ALP, Kao H‐Y, eds. Advances in Knowledge Discovery and Data Mining. Cham: Springer International Publishing; 2014:40–52.
Shah KS, Xu H, Matsouaka RA, Bhatt DL, Heidenreich PA, Hernandez AF, et al. Heart failure with preserved, borderline, and reduced ejection fraction: 5‐year outcomes. J Am Coll Cardiol 2017;70:2476–2486. doi:
Conrad N, Judge A, Canoy D, Tran J, Pinho‐Gomes A‐C, Millett ERC, et al. Temporal trends and patterns in mortality after incident heart failure: A longitudinal analysis of 86 000 individuals. JAMA Cardiol 2019;4:1102–1111. doi:
Farré N, Vela E, Clèries M, Bustins M, Cainzos‐Achirica M, Enjuanes C, et al. Medical resource use and expenditure in patients with chronic heart failure: A population‐based analysis of 88 195 patients: Medical resource use and expenditure in patients with chronic HF. Eur J Heart Fail 2016;18:1132–1140. doi:
Groenewegen A, Rutten FH, Mosterd A, Hoes AW. Epidemiology of heart failure. Eur J Heart Fail 2020;22:1342–1356. doi:
Nozaki A, Shirakabe A, Hata N, Kobayashi N, Okazaki H, Matsushita M, et al. The prognostic impact of gender in patients with acute heart failure—An evaluation of the age of female patients with severely decompensated acute heart failure. J Cardiol 2017;70:255–262. doi:
Simpson J, Jhund PS, Lund LH, Padmanabhan S, Claggett BL, Shen L, et al. Prognostic models derived in PARADIGM‐HF and validated in ATMOSPHERE and the Swedish Heart Failure Registry to predict mortality and morbidity in chronic heart failure. JAMA Cardiol 2020;5:432–441. doi:
Solomon SD, Dobson J, Pocock S, Skali H, McMurray JJV, Granger CB, et al. Influence of nonfatal hospitalization for heart failure on subsequent mortality in patients with chronic heart failure. Circulation 2007;116:1482–1487. doi:
Zoni‐Berisso M, Martignani C, Ammendola E, Narducci ML, Caruso D, Miracapillo G, et al. Mortality after cardioverter‐defibrillator replacement: Results of the DECODE survival score index. Heart Rhythm 2021;18:411–418. doi:
Køber L, Thune JJ, Nielsen JC, Haarbo J, Videbæk L, Korup E, et al. Defibrillator implantation in patients with nonischemic systolic heart failure. N Engl J Med 2016;375:1221–1230. doi:
Nevzorov R, Goldenberg I, Konstantino Y, Golovchiner G, Strasberg B, Souleiman M, et al. Developing a risk score to predict mortality in the first year after implantable cardioverter defibrillator implantation: Data from the Israeli ICD Registry. J Cardiovasc Electrophysiol 2018;29:1540–1547. doi:
Altibi AM, Prousi G, Agarwal M, Shah M, Tripathi B, Ram P, et al. Readmission‐free period and in‐hospital mortality at the time of first readmission in acute heart failure patients—NRD‐based analysis of 40,000 heart failure readmissions. Heart Fail Rev 2021;26:57–64. doi:
Setoguchi S, Stevenson LW, Schneeweiss S. Repeated hospitalizations predict mortality in the community population with heart failure. Am Heart J 2007;154:260–266. doi:
Rao A, Kim D, Darzi A, Majeed A, Aylin P, Bottle A. Regional variations in trajectories of long‐term readmission rates among patients in England with heart failure. BMC Cardiovasc Disord 2019;19:86. doi:
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2024. This work is published under http://creativecommons.org/licenses/by-nc-nd/4.0/ (the "License"). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
Aims
The primary objectives of this study were to analyse the nationwide healthcare trajectories of heart failure (HF) patients in France, 2 years after their first hospitalization, and to measure sequence similarities. Secondary objectives were to identify the association between trajectories and the risk of mortality.
Methods and results
A retrospective, observational study was conducted using data extracted from the Echantillon Généraliste des Bénéficiaires database, covering the period from 1 January 2008 to 31 December 2018. Follow‐up concluded upon death or at the end of the study. We developed a methodology specific to healthcare data by extracting frequent healthcare trajectories and measuring their similarity for use in a survival machine learning analysis. In total, 11 488 HF patients were included and followed up for an average of 2.9 ± 1.3 years. The mean age of the patients was 78.0 ± 13.2 years. The first‐year mortality rate was 31.7% and increased to 78.8% at 5 years. Fifty per cent of patients experienced re‐hospitalization for reasons related to cardiovascular diseases. We identified 1707 hospitalization sequences, and 21 sequences were associated with survival, while 15 sequences were linked to mortality. In all our models, age and gender emerged as the most significant predictors of mortality (permutation feature importance: 0.099 ± 0.00078 and 0.0087 ± 0.00018, respectively; weights could be interpreted in relative terms). Specifically, the age at initial hospitalization for HF was positively associated with mortality. Gender (male: 49.5%) was associated with poorer prognoses. Healthcare trajectories, including non‐surgical device treatments, valve replacements, and atrial fibrillation ablation, were associated with a better prognosis (permutation feature importance: 0.0047 ± 0.00011, 0.0014 ± 0.000073, and 0.00095 ± 0.000097, respectively), except in cases where these invasive treatments preceded or followed hospitalization for cardiac decompensation. The predominant negative prognosis sequences were mostly those that included HF‐related hospitalizations before or after other‐related hospitalizations (permutation feature importance: 0.0007 ± 0.000091 and 0.00011 ± 0.000045, respectively).
Conclusions
We highlight the value of healthcare trajectories on frequent hospitalization sequences, mortality, and prognosis and indicate the necessary prognostic value of HF re‐hospitalization. Our work may be an essential tool for better identification of at‐risk patients in order to increase and improve personalized care in the future.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details
1 CEMKA, Bourg‐la‐Reine, France, Clinical Research Unit, CIC‐EC 1418, European Hospital Georges‐Pompidou, APHP, Paris, France
2 Clinical Research Unit, CIC‐EC 1418, European Hospital Georges‐Pompidou, APHP, Paris, France
3 Referral Center for Cardiac Amyloidosis, Mondor Amyloidosis Network, GRC Amyloid Research Institute and Cardiology Department, INSERM Unit U955, Team 8, Paris‐Est Creteil University, Hospital Henri Mondor, Val‐de‐Marne, Créteil, France
4 Univ Rennes, CIC 1414 INSERM, IRMAR, Mathematics Institute of Rennes CNRS, Rennes, France