Comparing health-related quality of life in modified Rankin Scale grades: 15D results from 323 patients with brain arteriovenous malformation and population controls

HRQoL was measured by the generic self-administered 15D instrument. It can be used both as a profile and as a single index score measure. The questionnaire includes 15 dimensions: mobility, vision, hearing, breathing, sleeping, eating, speech, excretion, usual activities, mental functioning, discomfort and symptoms, depression, distress, vitality, and sexual activity. For each dimension, the respondent chooses one of the five ordinal levels best describing his/her state of health at the moment (best value = 1; worst value = 5) [36]. The single index score (15D score) represents the overall HRQoL on a 0–1 scale (1 = full health, 0 = being dead) and the dimension level values reflect the goodness of the levels relative to no problems on the dimension (= 1) and to being dead (= 0). They are calculated from the questionnaire using a set of population-based preference or utility weights. Mean dimension level values are used to draw 15D profiles for groups. A change or difference in the 15D score of ± 0.015 is clinically important [2]. Further properties of the instrument are described at http://​15d-instrument.​net/​15d/​.

Patients with missing data were excluded from the analysis of the variable or dimension in question. Two patients (0.9%) had not filled in the entire 15D questionnaire. They were included only in the dimension level analyses for the dimensions they had answered. The 15D data for the general population came from the National Health 2011 Survey representing the Finnish population aged over 18. For this analysis, those individuals were selected, who were from the Helsinki University Hospital catchment area and in the age range of patients (n = 1350). This sample was weighted to reflect the age and sex distribution of the patients, separately for each mRS grade [19]. The equality of the mean 15D scores across mRS grades was tested with ANCOVA (age and sex standardized), followed by Bonferroni corrected post hoc tests. The profiles comparing mRS grades to each other were drawn based on the estimated means after age and sex standardization. The p values < 0.05 were considered statistically significant. Statistical analysis was performed using the SPSS for Mac statistical software version 25 (SPSS, Inc., Chicago, IL, USA). The study adheres to STROBE reporting guidelines.

We explored the previous 5 years (2015 January to 2020 August) of brain AVM follow-up studies on adult patients using search terms “AVM,” “arteriovenous malformation,” “mRS,” “modified Rankin Scale,” and “functional outcome” on MEDLINE/PubMed. The search was conducted on 2 August 2020. We included studies which were published in English, had used mRS dichotomization in outcome evaluation, and reported the mRS cut point, mean follow-up time, and sample size. We excluded studies on pediatric and elderly patients with AVMs, studies which did not report follow-up time, number of patients, studies which did not use dichotomization of mRS and if the sample size was smaller than 20 patients. These criteria were evaluated first based on abstracts; however, if uncertainty existed based on this, the article was pulled for full-text review. The detailed protocol of the search is illustrated in Fig. 1.

Patients in mRS 0 (n = 154, mean 15D score = 0.954, SD =  ± 0.060) had better total HRQoL compared to age- and sex-standardized general population (mean = 0.927, SD =  ± 0.028), p < 0.0001 (Fig. 2). Figures 3 and 4 present the profiles for mRS grades 0 and 1, respectively, compared to the general population using age and sex standardization. The mean total 15D score for patients in mRS 1 (n = 78, mean = 0.844, SD =  ± 0.100) was worse than that of the age- and sex-standardized population controls (mean = 0.927, SD =  ± 0.021), p < 0.0001 (Fig. 2).

All the mRS grades differed significantly in their mean total 15D scores, except mRS 1 (mean = 0.844, 95% CI = 0.826–0.859) and mRS 2 (mean = 0.838, 95% CI = 0.813–0.868). When comparing all the grades to one another, the only HRQoL dimension with distinct values for each mRS grade was mobility (Fig. 3). Patients in mRS 0 compared to mRS 1 patients differed in all other HRQoL dimensions except breathing and eating (Figs. 4 and 5). The total scores were significantly different in pairwise comparison, with index score of 0.954 (95% CI = 0.942–0.966) for mRS 0 patients and 0.844 (95% CI = 0.826–0.859) for mRS 1.

All the 17 AVM follow-up studies using mRS dichotomization published within the previous 5 years are represented in Table 2. Nine studies (52.9%) categorized favorable outcomes as mRS 0–2 and unfavorable as mRS 3–5 [12, 15, 18, 20, 21, 23, 28, 37, 40]. The rest (47.1%) used the lower cut point of mRS 1 [6, 14, 16, 17, 24, 30, 31, 35]. All studies with high grade or brainstem AVMs used the higher cut point [12, 18, 21]. In the studies using the cut point of mRS 2, the mean follow-up time was 3.4 years (SD =  ± 3.1 years) and for the cut point mRS 1 studies 2.4 years (SD =  ± 1.9 years).

We observed that in the long-term follow-up, patients in mRS 0 had a considerably better total HRQoL when compared to patients in mRS 1. To strengthen this point, patients in mRS 0, unlike those in all the other grades, had even better subjective total HRQoL than the age- and sex-standardized general population. Patients in mRS 1 and mRS 2 had a very similar HRQoL profile, even though they are often characterized as being very different by setting the mRS cut point between them, as delineated by the literature review. The higher cut point between 2 and 3 was equally common in modern AVM research as the lower one, and all the studies with worse expected results used the higher cut point [12, 18, 21].

According to our results, patients in mRS 0 formed a distinctive group with superior HRQoL compared to the rest. This finding is against the recent findings of two meta-analyses using the EQ-5D instrument on stroke patients [32, 41]. In these studies, patients in mRS 0 and 1 formed the most closely associated grades according to utility weights. The contradicting finding of our study could be owing to the different HRQoL instruments used. 15D has in comparison to EQ-5D and many other commonly used instruments illustrated a better detection rate for the psychological dimensions [27]. Furthermore, in comparison to EQ-5D, 15D has a lower ceiling effect and a higher detection rate for change [13]. Having a high ceiling effect means that the improvement in HRQoL with patients in the best possible health states might remain undetected because the scale runs out. Thus, EQ-5D could overreact to better than average health states by producing full index scores too easily [13]. Our findings could also illustrate a difference in the pathophysiological nature of the diseases between the mostly ischemic stroke patients in the aforementioned meta-analyses and the AVM patients in our study. Regarding the comparison to the general population, it should be noted that a random population sample contains participants with worse functional capability than mRS 0, which could explain some of the excellent HRQoL values of the mRS 0 AVM patients. Also, it might be difficult to control individuals to estimate their common symptomology compared to patients who have possibly experienced real disabling symptoms. However, even with this, it is fair to say that mRS 0 patients have at least similar subjective HRQoL with the general population or possibly even better.

Modified Rankin Scale dichotomization was first applied to an acute stroke trial in the NINDS (National Institute of Neurologic Diseases and Stroke) tissue plasminogen activator trial in 1995, in which the grade was divided into favorable outcome (mRS 0–1) and unfavorable outcome (mRS 2–5) [11], Afterwards, dichotomization has become common also in other neurological and neurosurgical studies, with an ongoing debate about the rightful dichotomy cut point [10]. Between 2007 and 2016, more than half of the published stroke studies had used dichotomy in their statistical analyses [26]. Dichotomization has some statistical advantages [1]. It can lower the error rates of interobserver variability, especially in the mid-range of the scale, and it eases the reporting and interpretation of the results [22]. However, the cut point needs to be carefully chosen. It should support the severity of the illness and the point the effect is anticipated [7]. If the dichotomization cut point is set incorrectly, it can distort the results as it hides the distribution of the individual grades inside the dichotomous class. This loss of information, however, always exists with dichotomization, unrelatedly of the correctness of the cut point [9]. Often, the cut point is chosen to be set at the sample median. Despite often regarded to be an increasing factor of statistical power, a so-called median split often actually reduces statistical power and can lead to falsely significant results [3]. To avoid these issues, numerous non-dichotomic statistical methods have been developed which rather than compare two fixed classes to one another, attempt to better take into account the movement across the whole scale [5, 25]. These ordinal analyses of mRS have proved more reliable than dichotomy in reporting the outcomes and cost-analyses of stroke patients [10, 34]. Even though these alternative methods are available, dichotomization exists in our research and as many hallmark studies have used it, a deeper understanding of the dichotomy and the individual grades can improve our interpretation of the existing literature. In the review of AVM studies from the past 5 years, the mRS cut points were equally either between mRS 1 and 2 or between 2 and 3. Interestingly, the studies which used the higher cut point were those with difficult AVMs and supposedly worse outcomes. It is tempting to speculate on the possible reasons behind the decision of a higher cut point. With a worse expected outcome of patients with for example brainstem AVM compared to cortical AVMs, it seems reasonable to choose a higher cut point. Is this reason because with the worse expected results we can loosen the definition of a favorable outcome or because of the inability to catch statistically significant results without a higher cut point?

Dividing outcomes into “good” and “bad” is problematic. A good outcome for some patients might represent unfavorable outcomes for others. For example, after a devastating AVM bleeding leading to severe disability, an improvement in functioning somewhat independently might be perceived as a favorable outcome. In contrast, for a preoperatively asymptomatic AVM patient, a postoperative development of minor symptoms might appear as an unfavorable outcome. As this phenomenon is based on the patient’s subjectivity to the symptoms, it should not substantially affect the evaluation of the traditional mRS, which is an objective functional outcome instrument. Objectivity is needed in research, as it allows transparent comparison of treatment strategies, patients, and institutions. However, when thinking about the quality of care, we cannot be content with only the objective outcome. To understand how the objective mRS translates into HRQoL, we compared the mRS with the 15D results. The differences between the mRS grades were smallest in the psychological dimensions and the biggest differences were illustrated in the dimensions requiring physical capability. This inability to discriminate between the psychological dimensions could be owing to the lack of statistical power; however, in our previous HRQoL report, we were able to distinguish significant differences between the certain subgroups of AVM patients regarding their mental well-being using the 15D and the same patient population, although with smaller sample size [29]. Given that the traditional mRS is a scale of functional outcome and the ability to continue previous usual activities, it was able to illustrate these also in the subjective HRQoL results, apart from the overlapping in mRS 1 and 2 in the ability to continue previous usual activities. Despite being able to illustrate both objective and subjective functional outcomes, the division of good and bad mRS remains difficult. As the goal of treatment is often to improve the patient’s condition or to prevent or stop it from declining, it would seem logical to also set our methodology so that this could be captured, instead of trying to artificially divide the scale into favorable and unfavorable.

In the history of mRS dichotomization, two common alternatives exist [10]. First, patients in mRS 1 can still carry on with their previous activities, whereas in mRS 2, they cannot, reasoning the cut in between them [10]. On the other hand, mRS 2 patients can still look after their own affairs without assistance when compared to mRS 3, and therefore, mRS grades ≤ 2 are defined to indicate functional independence, giving an alternative cut point [4, 10]. According to our literature review, both were equally used in neurosurgical studies. One of the dimensions of the HRQoL questionnaire used in our study measures the ability to continue previous usual activities. According to our results, there was no considerable difference in this dimension between the AVM patients in mRS 1 and mRS 2. Also, when comparing mRS 1 AVM patients to the population, there was a serious drop in the ability to continue previous activities, even though by definition they should be able to carry on with their normal lifestyle. These findings support the cut point for functional dependence between the grade mRS 2 and 3. When considering the massive drop in the ability to continue previous activities between mRS 0 and mRS 1 patients, however, it would seem more reasonable to consider mRS 0 patients separately from the rest. Including them in the same dichotomous group with mRS 1 patients could lead to a considerable bias by improving results for mRS 1 patients and worsening them for mRS 0 patients. This effect always exists with dichotomization; however, the magnitude of it in our study sample was rather extreme, as illustrated by the comparison of the patients to the general population. Regarding the ordinal mRS analyses, it should be noted that the gap of HRQoL between mRS 0 and mRS 1 could be substantially greater than the gap between the remaining grades. As already discussed, these findings differ from the UW-mRS stroke studies which regard these grades close in utility. A subject of interest for future research is to differentiate whether these findings are owing to the different HRQoL instruments used or the differences in the patient populations.