Determinants of Quality of Life in Breast Cancer: Meta-Analytic Structural Equation Modeling of Studies

,


Introduction
Breast cancer (BC) has the highest incidence and mortality among females, with an estimated incidence of 24.2% of all cancer types in 2018 worldwide [1].Te estimated 5-year survival rate for women with breast cancer is 80-90%, with poor rates in advanced stages [2].Hence, enhancing the quality of life (QoL) in these patients is of high importance.
Despite remarkable achievements in control of the disease, nausea, vomiting, pain, insomnia, anorexia, and fatigue are common treatment side efects in patients with BC that result in psychosocial problems and lower activity and worsened QoL [3,4].It is claimed that poor QoL is associated with shorter survival, lower treatment adherence, increased cancer mortality, longer hospital stays, and reduced self-care [5,6].Some research introduces QoL as a prognostic factor with impacts comparable to pharmacological treatments [7].However, there is no consensus on the defnition of QoL as a multidimensional subjective phenomenon that includes all physical and emotional aspects [8].Te World Health Organization (WHO) defnes QoL as "Te situation of life resulting from the mixture of the impact of a large number of factors such as those infuenced on happiness including being in comfort physical environment, satisfying occupation, intellectual and social attainments, justice, freedom of actions, expression, and also the health aspects" [9].
Irrespective of disease stage and type of treatment, QoL of patients with BC is afected through changes in fatigue, physical inactivity, sleep disorder, and psychological distress immediately after diagnosis [10,11].Several studies have assessed determinants of QoL in BC.However, the number of included factors in each study is limited, and the results are sometimes contradictory.No comprehensive study has been conducted to consider a large set of factors in a coherent causal network.Te current study aimed to evaluate the impact of the most critical factors in QoL of patients with BC by using a meta-analytic structural equation modeling approach.From the factors assessed in the literature, we selected variables with non-ignorable evidence that includes body mass index (BMI), physical activity (PA), sleep, depression (Dep), stress, and fatigue.

Materials and Methods
2.1.Literature Review and Data Extraction.We searched PubMed, Scopus, Cochrane, and Web of Science databases for relevant published papers with the combination of keywords and specifc terms as follows: (BMI OR depressive OR Physical activity OR Sleep OR Fatigue OR Mood OR stress) AND (Quality of Life) AND (Breast Cancer Survivors OR Breast Cancer OR Neoplasm).Te details of the diferent search strategies are provided in the online resource materials (search queries).We also reviewed the reference lists of the original articles and reviews to identify other potentially eligible papers.Studies meeting the following criteria were included in the study: (1) being conducted on BC survivors, (2) written in English, and (3) reporting correlation coefcient between variables such as BMI, depression, physical activity, sleep, stress, fatigue, and the quality of life, directly or indirectly.Te quality of life, stress, depression, sleep quality, and physical activity were measured by different related questionnaires.Details of the primary data are provided in the online resource material (Table 1: efect size).
Te full text of potentially relevant articles was obtained.Two authors extracted data independently by using a form based on the Cochrane Collaboration's data extraction rules, and a third author resolved any discrepancies in the evaluation of the studies.Preferred Reporting Items for Systematic Reviews and Meta-Analysis for Protocols 2015 (PRISMA-IP 2015) were used for preparation and reporting [12].
A two-stage meta-analytical structural equation model (MASEM) was applied to test the relationship between QoL and other components between breast cancer survivors.In this approach, we frst pooled the multiple correlation matrices available in the studies by meta-analysis, and then the relations were analyzed using structural equation modeling.Various methods have been proposed in the literature to pool the correlation coefcients.In this paper, a two-stage approach synthesizes covariance matrices in meta-analytic structural equation modeling to test the power of correlations between components [13].Te method used in this paper was based on two stages, MASEM.In the frst stage, the correlation matrices were tested for heterogeneity assumption.If they were homogeneous, they were combined to a pooled estimate.If there was no homogeneity, the random efect meta-analysis approach was used [13].In the second stage, we run the SEM with combined efect sizes [13].Considering moderator, we should correlate the direct and indirect efects in searched paper to combine the efects by meta-analysis.Unfortunately, there were not enough in the papers for considering moderators.When the criteria presented in diferent articles to evaluate the treatment under study are diferent (regression beta coefcients, odds ratio, chi-square statistic, F statistic, and Z statistic), they should be transferred to correlation coefcient as the same efect size.If the beta regression coefcient was reported in a study, it transferred to correlation coefcients under the condition that the beta coefcient measure was a number between ±0.5 [14].If the efect size is reported as OR, it could be converted to the correlation coefcient according to the formula r � (OR 3/4 − 1)/(OR 3/4 + 1) [15].For Z statistic computed from testing of equality of two population means, correlation is calculated as r � Z/ �� N √ [16].N is the total study sample size.Also, F statistic is computed from ANOVA test, and correlation is calculated as r � ����������� (F/(F + df))  [16].Te following items were extracted for each study: frst author name, year of publication, sample size, primary goal of the study, the correlation between variables, mean age, and the nationality or the race of the participants in the study.Studies with missing or unrelated information were deleted.Te initial literature search produced 5238 potentially relevant studies, from which 1051 and 3505 studies were removed for being, respectively, duplicate or irrelevant.Tis led to 73 studies relevant for inclusion in fnal analysis.Te process of article selection is shown in Figure 1.Te number of reported correlation was 3-18.None of included articles provided all correlations between variables.Egger's test was used to evaluate publication bias [17].

Critical Appraisal: Assessment of Study
Bias.Te quality of relevant articles was evaluated using the Newcastle-Ottawa Scale (NOS) for cohort studies [18].Studies were evaluated based on exposure, comparability, selection, and outcome.Te maximum possible score (least risk of bias) was nine stars.Moderate to good quality was determined by scores of fve stars or more [18].
Te Jadad scale was used for quality assessments of the randomized clinical trials (RCTs) [19,20].Tis scale comprises fve questions related to the validity of RCTs.Te total scores range from 0 to 5 points, where trials with 0-2 points are considered poor quality, where a score of 3-5 denotes a high-quality RCT [20].
Te assessment and scoring system is provided in the online resource (assessing the quality).Two review researchers independently evaluated the fndings of each study to confrm an unbiased evaluation.

Statistical Analysis.
Atwo-stagemeta-analytical structural equation model (MASEM) was applied to test the relationship between QoL and other components between breast cancer survivors.In this approach, we frst pooled the multiple correlation matrices available in the studies by meta-analysis, and then the relations were analyzed using structural equation modeling.We tested the homogeneity of the correlation 2 European Journal of Cancer Care matrices from individual studies using I 2 and Q statistic.I 2 values above 75% indicate serious heterogeneity where values lower than 25% showed minor heterogeneity.p value <0.05 indicates heterogeneity among studies and the need to use a random efect model [18].Te null hypothesis for the Q test also declares homogeneity [18].Ten, a weighted pooled correlation matrix was calculated.To build a pooled matrix, the patterns of correlations between independent and response variables need to be fairly similar in diferent studies.A key issue is choosing a fxed or random efect model based on the study target [21].In fxed efect models, the size of the actual efect is shared in all studies.In contrast, in random efect models, efect sizes are assumed to difer among studies and are usually assumed to follow a normal distribution [22].MASEM provides standardized path coefcients and tests the correlations between components based on the following goodness-of-ft criteria with desired ranges in parentheses: root mean square error of approximation (RMSEA < 0.06), comparative ft index (CFI > 0.95), standardized root mean square residual (SRMR < 0.08), and TLI index [23].

Results
Most of the retrieved eligible studies (55 of 73) were dated to 2010 and were conducted in the United States of America.
Among the searched databases, PubMed had the most relevant articles.Te I 2 values for the assessed correlations ranged from 14.31 to 98.96%, and the Q test had a value of less than 0.001 in most cases, both indicating high heterogeneity among studies.Hence, we adopted the random efect model in this study.
Our model assumed the following correlation between variables: sleep with BMI and PA; Dep with BMI, PA, and sleep; fatigue with sleep and Dep.; and stress with PA, Dep, BMI, fatigue, and sleep (Figure 2).Te χ 2 value for this model was 4.24, with a p value of 0.23, indicating a good ft.RMSEA and SRMR values were 0.003 and 0.0312, respectively, which confrm the suitability of the model.Te TLI value of 0.97 and the CFI value of 0.99 indicate an acceptable ft of the fnal model (Table 1).In structural equation modeling, the degree of freedom is calculated from the df = 0.5 × (p) × (p + 1) − k formula.P is the number of observable variables, and k is the number of parameters that the software will calculate in the model.Tis model has six obvious variables, so six factors and six measurement errors are calculated for the model.Also, the six coefcient paths must be calculated.So, 6 + 6 + 6 is equal to 18. Ten, the was 3 (df = 0.5(6) (6 + 1) − 18 = 3).Table 2 in online resource materials shows the summery of data from studies included in the fnal analysis.For each efect, the value and 95% confdence interval for the merged correlations are provided.Te largest correlation was between Dep-Stress (0.62, 95% CI = 0.4748; 0.7468), Fatigue-Dep (0.47, 95% CI = 0.3453; 0.5922), Stress-Fatigue (0.45, 95% CI = 0.1463; 0.6875), PA-  2).Te quality of included studies was also evaluated and presented in the online resource materials (Tables 3 and 4) According to the results of MASEM (Figure 2), the highest positive efect on QoL was for PA (path coefcient = 0.33, 95% CI = −0.0444;0.6334), where fatigue (path coefcient = −0.23,95% CI = −0.6825;0.0361) and stress (path coefcient = −0.22,95% CI = −0.5143;0.6875) had the most detrimental efect on QoL in patients with BC.In this model, approximately 68% of QoL variance is determined with the variables included in the model.Tere was no publication bias according to Egger's test (P � 0.78) (Table 3).

Discussion
In the present study, we assessed the efect of various factors on QoL in BC survivors by using meta-analytic structural equation modeling using 73 studies from the literature.We should say that two issues would be considered in the way of selecting the input variables.
(1) Tere were a sufcient number of articles on that variable.(2) Te current approach (MASEM) presented by extracting the correlation between diferent variables from study units then pooling the multiple correlation matrices available in the studies by meta-analysis, secondly the measure of relations were analyzed using structural equation modeling.
According to these descriptions, having the correlation coefcient information on that variable with QoL and each other predictor was the second criterion for considering that variable in modeling.
We had no chance of considering more predictors or presenting diferent moderators in our model for these two reasons.
In this paper, we used a correlation-based MASEM model.According to these issues, the mean age was insufcient to calculate the correlation efect size as the model input.On the other hand, the latent efect of age or the cancer stage can be seen in physical activity, fatigue, or other predictors.
Our fndings highlight the signifcance of the physical activity, stress, and fatigue in this regard and the results indicate that the null hypothesis (equality of regression coefcients across predictors) was rejected and the efect of variables was not the same (x 2 � 32.7564, p value < 0.001, R � 0.45).It is noteworthy that most studies have been conducted over the past ten years, which underscores the increasing concerns about the QoL in patients with breast cancer in recent years.QoL is also recognized as a signifcant predictor of prognosis in cancer patients [25].However, to our knowledge, no research has assessed determinants of QoL in this population from accumulated data thus far, and all original research  European Journal of Cancer Care studies have focused on a very few factors with contradictory results in some cases.Despite the improvements in the treatment of BC and the increasing number of survivors, there is currently no study that comprehensively addresses the factors afecting the quality of life of these people, and in all studies, only one or two aspects of these factors have been mentioned.Also, the results obtained in these studies are sometimes contradictory or diferent.Terefore, future research should examine the quality of life in all aspects of interaction with other variables.Te strengths of this study are the use of both meta-analysis and structural equation modeling and cumulating the results of other studies.According to the ftting values of the model, it is evident that these models are largely satisfactory and represent the factors afecting the quality of life.

Study Limitation.
Only papers published in English were included in this study.

Clinical Implication.
QoL is the crucial factor in breast cancer survivors.Enhancing physical activity and reducing fatigue and stress could improve QoL in patients with breast cancer.

Conclusion
Findings of the current meta-analytic study indicate that physical activity is critical in enhancing the quality of life in patients with breast cancer.Controlling fatigue and stress is of high importance and maintains a high quality of life in these patients.Further large-scale studies are essential to

Figure 1 :
Figure 1: Flow of the study selection process based on PRISMA guideline.

Figure 2 :
Figure 2: Determinants of quality of life (QoL) in breast cancer survivors using meta-analytic structural equation modeling.BMI, body mass index; DEP, depression; PA, physical activity.

Table 2 :
Determinants of quality of life in patients with breast cancer by using meta-analytic structural equation modeling.
QoL, quality of life; BMI, body mass index; PA, physical activity.

Table 3 :
Te pooled correlation matrix from stage 2, along with the heterogeneity.

Table 1 :
Goodness-of-ft indices for meta-analytic structural equation modeling.European Journal of Cancer Care fortify these fndings, fnd other vital factors, and assess their interrelations.