Quality of life among elderly patients undergoing transcatheter or surgical aortic valve replacement– a model-based longitudinal data analysis

The TAVI Calculation of Costs Trial (TCCT), which this study is based on, was designed as a prospective observational multicenter cohort study on elderly patients with symptomatic AS receiving either SAVR, TAVR, or best medical therapy (DRUG). This study has been approved by the institutional ethics committee (Research Ethics Committee Albert-Ludwigs-Universität Freiburg, Germany ID: 52/11), and registered in the German Clinical Trial Register (ID: DRKS00000797). All patients referred to our centers between April 2011 and October 2013 were considered for inclusion into the study. Age above 75 years was deliberately chosen as an inclusion criterion. All treatment decisions were made by a study-independent “heart team” of cardiac surgeons and cardiologists according to best clinical practice [13].

A primary focus of the study was to evaluate QoL before and after procedure as well as during a 2-year follow-up period using the EQ-5D questionnaire, a standardised instrument to measure health outcome whose design emphasizes simplicity and quick completion, improving its feasibility and thereby response rates from this study’s severely health-impaired patient population [14]. EQ-5D_Index values were calculated using the established German calculation formula which is based on time trade off values [15]. A total of 2463 questionnaires were completed by 169 elderly patients treated with either transcatheter aortic valve replacement (TAVR, N = 92, of which 30 received TAVI via the transapical approach), surgical aortic valve replacement (SAVR, N = 70, of which 17 received additional bypass grafting and/or mitral valve replacement), or drug-based therapy (DRUG, N = 7), who were tracked using monthly telephone interviews. Cardiovascular and non-cardiovascular complications, prosthetic valve associated endpoints, and therapy-specific endpoints were assessed according to the revised definitions provided by the Valve Academic Research Consortium (VARC-2) [16].

Differences between treatment groups were analyzed using the non-parametric Wilcoxon rank-sum test. Linear and ordered logistic regression analyses with a random intercept were carried out for the EQ-5D questionnaire's EQ-5D_Index (continuous endpoint, 0 = worst state, 1 = best state), the 5 EQ-5D dimensions (order categorical endpoint, 1 = no problems, 2 = some problems, 3 = extreme problems), patients heart function according to the New York Heart Association (NYHA) Functional Classification and the Canadian Cardiovascular Society (CCS) Angina Grading Scale. In addition, the current status of assistance at follow-up was retrieved and an order categorical variable was built with the following characteristics: 1 (no external help or intra-familial help only), 2 (assisted living and/or home care services) and 3 (short or long-term care or temporal hospitalization). Distance to death was modeled as a categorical variable for the last six months before death. With respect to EQ-5D_Index values during follow up, overall measurement compliance was 56.6 %, with ~85 % compliance during the first three months, ~80 % compliance during the first six months, and ~73 % compliance during the first year, but only ~40 % compliance during the second year. In addition, there are a number of missing values with respect to EQ-5D dimensions, patients heart function, their current status of assistance and clinical complications. With respect to these missing values no imputation method was applied.

All analyses were performed using Stata 14 (Stata Corp., Texas. USA).

In Fig. 1, the results of a linear regression analysis with a random intercept at the patient level and observational time (in moths) as a categorical covariate are shown.

Over the two-year study period, three main effects were observed. Firstly, QoL measures decreased slightly over time. Secondly, QoL measures were substantially impaired at month 1 after the initial episode of hospitalization. In addition, 39 cases of mortality were reported during follow-up. Of the 39 cases of mortality, 20 % died during the first month post discharge, 33 % died during the first three months post discharge, 49 % died during the first six months post discharge, 78 % died during the first year post discharge and 90 % died during the first 18 months post discharge. QoL decreased substantially at the end of life with a measurable amount starting at the sixth from last follow-up (month) before death.¹ Table 2 summarizes these three effects on EQ-5D_Index values and also shows to which degree the five EQ-5D dimensions (model (2) to (6)) and measures of heart function (model (7) and (8)) are affected by these effects. At the last follow-up before death, for instance, EQ-5D_Index values are most impaired by an average of −0.495 (p < 0.001) irrespective of time- and post-discharge effects. The same is true for all 5 EQ-5D dimensions (see model (2) to (6) in Table 2) as well as measures of heart function (see model (7) and (8) in Table 2) and the current state of assistance (see model (9) in Table 2). In addition, the current state of assistance is target to a monthly trend, meaning that the risk for worsening of the status of assistance increases by every month of the two year study period (OR 1.06, p < 0.05). The same is true for the EQ-5D dimensions of mobility and pain as well as the heart function measured by the NYHA Functional Classification. At month 1 after the initial episode of hospitalization, nearly all QoL measures were substantially impaired, which is also true for the NYHA Functional Classification (OR1.94, p < 0.01) and the current status of assistance (OR 2.65, p < 0.01). Although both consist of four categories, the NYHA functional classification seems to be more sensitive to trend, post-discharge and time-to-death effects than the CCS grading scale.

According to the European Society of Cardiology/European Association for Cardio-Thoracic Surgery (ESC/EACTS) Guidelines [17], TAVR is recommended for patients considered unsuitable for conventional surgery because of severe comorbidities. In patients considered to be at high risk for conventional surgery, these comorbidities and the associated individual patient’s risk should be assessed by a “heart team” of cardiac surgeons and cardiologists to select the most optimal treatment strategy for individual patients [17]. As a result of this modus operandi, treatment decision is associated with systematic risk selection and the logistic EuroSCORE has previously been shown to be of relevance for predicting both treatment decisions and overall survival [13]. Interestingly, when predicting follow-up QoL measures rather than survival, baseline values of the respective QoL measure had a more dominant impact on future values of any of the observed measures than risk profiles in comparison to the logistic EuroSCORE (see Table 3, where values of the dependent variables at baseline and EuroSCORE values at baseline were included as time-invariant variables). Interestingly, the impact of baseline values on the respective dependent variable seems to persist over the entire two-year period.²

Furthermore, a number of between-group differences may be observed. In comparison to TAVR patients, SAVR patients report more favorable QoL measures (see model (1) to (6) in Table 3), and, (or maybe because of this), a reduced risk for worsening of the status of assistance is recorded for them (OR 0.0441, p < 0.001). On the other hand, an increased risk for limitations in the heart function is recorded for the few patients in the Drug group (see model (7) and (8) in Table 3). Please note that these between-group differences in QoL measures should not, or at least only in part, be interpreted as treatment effects, as the different treatment groups are not randomly assigned but subject to a risk-driven patient selection which may not be fully addressed by the applied baseline-adjustment.

As shown in Table 1, a number of clinical events were recorded for TAVR and SAVR-patients, of which nearly 80 % occurred during the initial episode of hospitalization. In order to analyze the impact of such events in the short and medium term, Drug patients were excluded from the analyses. As shown in model (1) in Table 4, QoL is most affected by life threatening non-access site bleeding (one-month change: −0.447, p < 0.001), stroke (one-month change: −0.161, p < 0.05) and stage 3 acute kidney injury (one-month change: −0.177, p < 0.01). Interestingly, only life threatening non-access site bleeding was associated with a more intense status of assistance. In a next step, we analyzed the medium term impact of the respective complications. Therefore, categorizations of variables included in Table 5 were restructured by pooling complications of minor interest in ‘other complications’. Then, lagged variables were included in order to capture the medium term impact of the respective complications. As shown in Table 5 model (2), time lags of up to three months were included in the analyses³. Interestingly, the impact of lagged effects on EQ-5D_Index values remained significant for life threatening access site bleeding only. Other complications, such as life threatening non-access site bleeding (non-ASB), stroke, and stage 2 and stage 3 acute kidney injury (AKIN 2 and 3), are still associated with substantial effects impairments of EQ-5D_Index values at the first follow-up interview after event, but the event-related impairments decrease dramatically at the second and third follow-up interviews after event. In Table 5 we also consider the impact of two different strategies to take the influence of the distance to death into account: We may just exclude measurements from the last six months prior to death (models (3) and (4)) or we may just ignore the distance in the modelling (models (5) and (6)) and equal EQ-5D_Index to 0 in the month of death. We can observe similar results for 3 complication groups, but changes for non-ASB and AKIN 3. For non-ASB this seems to be related to the fact that this complication took place in 12 patients, of which two died within three months after event. These two patients were facing the worst possible health state EQ-5D_Index = −0.205 during each follow-up interview after event and before death. Unsurprisingly, the estimated impact of life threatening access side bleeding changed substantially, as these two patients were excluded from the analysis in model (3) and (4). Basically, the results of model (1) to (4) represent the estimated impact of the respective complication on EQ-5D_Index values assuming that the patient survives the six months following the event. Model (5) and (6), in contrast, show the estimated impact of the respective complication on EQ-5D_Index values for all patients, including those patients who die shortly after the event. That is why, for instance, model (6) returns substantially higher QoL impairments for patients with AKIN than model (2): Model (2) separates prior-to-death impairments from those arising in patients that do not die during the next months, whereas model (6) does not adjust for upcoming death. Of the sixteen patients facing stage 3 AKIN, nine patients died before the first follow-up after and two patients died during month two and three after event.