- Journal List
- HHS Author Manuscripts
- PMC3665161
As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsem*nt of, or agreement with, the contents by NLM or the National Institutes of Health.
Learn more: PMC Disclaimer | PMC Copyright Notice
Leuk Lymphoma. Author manuscript; available in PMC 2013 May 28.
Published in final edited form as:
Leuk Lymphoma. 2013 Feb; 54(2): 290–297.
Published online 2012 Sep 6. doi:10.3109/10428194.2012.711830
PMCID: PMC3665161
NIHMSID: NIHMS469858
PMID: 22799432
KJ Yost,1 CA Thompson,2 DT Eton,1 C Allmer,1 SL Ehlers,3 TM Habermann,2 TD Shanafelt,2 MJ Maurer,1 SL Slager,1 BK Link,4 and JR Cerhan1
Author information Copyright and License information PMC Disclaimer
The publisher's final edited version of this article is available at Leuk Lymphoma
Abstract
Quality of life (QoL) is an important outcome in patients with non-Hodgkin lymphoma (NHL). We assessed the validity of administering the Functional Assessment of Cancer Therapy – General (FACT-G) at 12-month intervals over 3 years in a longitudinal study of 611 prospectively enrolled, newly diagnosed NHL patients. We evaluated corrected item-total correlation and percent missing to identify items that may be less useful in certain NHL patient subgroups. The FACT-G subscales and total score demonstrated good internal consistency reliability, convergent validity and known-groups validity. Most scores also demonstrated good responsiveness to change. Questions that could be problematic included GE3 (losing hope) and GP2 (nausea) for patients in remission, and GP5 (bothered by side effects) for patients being observed. Overall, the FACT-G was a valid measure for monitoring QoL over time in NHL patients. However, sensitivity analyses based on subscale scoring that excludes potentially problematic items may be warranted.
Keywords: non-Hodgkin lymphoma, quality of life, validation studies, patient-reported outcomes
INTRODUCTION
Incidence rates for non-Hodgkin lymphoma (NHL) in the U.S. have been stable since 1991 for men and have been increasing by 1.1% per year in women since 1990. Mortality rates in the U.S. for non-Hodgkin lymphoma have been decreasing by 3.0% in men and 3.7% in women since 1997, and the overall 5-year survival is 67% [1]. This combination of stable or slightly increasing incidence rates and decreasing mortality rates means that more men and women are living with NHL. As a result, improving quality of life (QoL) has become an increasingly important goal in the management of patients with NHL.
NHL is a heterogeneous disease, with over 70 subtypes recognized by the World Health Organization [2]. The clinical course for NHL patients is varied and depends on a variety of factors, with NHL subtype, clinical and laboratory factors (e.g. age, stage of disease, performance status, lactate dehydrogenase level, and other subtype-specific factors) being the strongest predictors [3]. Patients with indolent or low-grade subtypes may be initially observed or treated with less aggressive approaches, while patients with aggressive subtypes are treated at the time of diagnosis, generally with chemoimmunotherapy. Some patients achieve long-term remission after initial therapy, while other patients have progressive or relapsing disease that generally requires treatment (active disease/treatment). These heterogeneous scenarios present challenges for monitoring QoL over time, as issues relevant to QoL are likely to differ with observation, disease activity or remission status. An ideal QoL measure should be general enough to be relevant to all situations, yet targeted enough to capture disease- and treatment-specific concerns.
In 2002, we initiated a prospective cohort study of NHL outcomes [4]. The Functional Assessment of Cancer Therapy – General (FACT-G) was selected as the measure of QoL in this cohort. The FACT-G was developed to measure QoL in cancer patients receiving therapy. While the FACT-G is widely used across diverse cancer patient populations, it was not initially developed to monitor QoL of patients over a longer follow-up period or for patients in remission or on observation. It was initially validated using classical test theory (CTT) methodology in a heterogeneous sample of 545 patients with cancer, 8% of whom had either leukemia or lymphoma [5]. We considered the SF-36 [6] when selecting the QoL measure, but determined it was too general. The FACT family of instruments follows a core plus module design. The FACT-G comprises the core set of questions, which can be answered by a broad range of patients. To this core, modules of questions that are specific to a particular type of cancer can be added. The FACT-Lymphoma (FACT-Lym) consists of the FACT-G plus 15 questions measuring symptoms or concerns specific to patients being treated for lymphoma. In 2002, the FACT-Lym was not yet available. Once the FACT-Lym became available, we reviewed the content of the 15-item lymphoma-specific subscale and decided to continue using the FACT-G rather than the FACT-Lym because the items in lymphoma subscale (e.g. “I have night sweats” and “I am bothered by itching”) [7] would not be applicable for many of the patients in the cohort over time. Specifically, the FACT-Lym was judged to be targeted for new patients on active treatment, as opposed to monitoring QoL in NHL patients on observation and in remission.
Our objectives were to (1) assess the validity of the FACT-G in a large sample of NHL patients, and (2) determine the validity of administering the FACT-G over time in NHL patients on observation, active disease or treatment (initial or at progression/relapse), and in remission.
METHODS
Patients
Patients were selected from the Molecular Epidemiology Resource (MER) of the University of Iowa/Mayo Clinic Lymphoma Specialized Program of Research Excellence (SPORE). The MER is a prospective cohort study of NHL outcomes that was initiated in 2002 [4]. Eligibility criteria included patients with NHL who were evaluated at Mayo Clinic Rochester and the University of Iowa within 9 months of their initial diagnosis, age 18 years or older, U.S. residency, and enrolled in MER from 2002-2008. Exclusion criteria included HIV infection, non-English speaking, and inability to provide written informed consent. At baseline, patients provided data on personal and family history, comorbidities, functional status and QoL. Patients were prospectively contacted every 6 months during the first 3 years after initial diagnosis to assess vital status, disease progression, and new treatments; QoL was assessed at 1-, 2- and 3-year follow-ups.
QoL Instrument
QoL was assessed using the FACT-G. The FACT-G is comprised of four subscales: physical well-being (PWB; 7-items, score range 0-28), social/family well-being (SWB; 7-items, score range 0-28), emotional well-being (EWB; 6-items, score range 0-24), and functional well-being (FWB; 7-items, score range 0-28). Users of the FACT-G are able to generate an overall score and four subscale scores with ranges and distributions that are sample-specific. All questions in the FACT-G use a 5-point rating scale (0 = Not at all; 1 = A little bit; 2 = Somewhat; 3 = Quite a bit; and 4 = Very much). Provided more than 50% of the items comprising a subscale are answered, a subscale score is computed as the prorated sum of the item responses for that subscale. Prorating, which replaces missing values with the mean of the completed items for that subscale, has been shown to be an acceptable method of imputing missing data in the FACT instruments when more than 50% of the items are answered [8]. The FACT-G total score is computed as the sum of the four subscale scores, provided the overall item response is at least 80% (i.e. at least 22 of the 27 items were answered) and has a possible range of 0-108 points. Negatively worded items are reverse scored prior to summing so that higher subscale and total scores indicate better QoL [9].
Statistical Analysis
Analytic sample
Of the 1,691 patients enrolled, 1,034 had one of the eligible NHL aggressive or indolent subtypes (Table I). Of these 1,304 patients, 611 patients had a baseline QoL assessment obtained on or before the date of initiation of their first treatment, and comprise the analytic sample. At 12 months, 434 (71.0%) participated, 35 (5.7%) were deceased, 14 (2.3%) had withdrawn, and 128 (20.9%) did not respond or did not respond in the allotted time. At 24 months, 422 (69.1%) participated, 46 (7.5%) were deceased, 18 (2.9%) had withdrawn, and 125 (20.5%) did not respond or did not respond in the allotted time. At 36 months, 231 (37.8%) participated, 68 (11.1%) were deceased, 19 (3.1%) had withdrawn, and 293 (48.0%) did not respond or did not respond in the allotted time (due to a delay in the 36 month assessment being added to the protocol). Treatment status (Observation, Active treatment/Progression, Remission) was determined at each assessment.
Table I
Patient characteristics at the baseline assessment (n=611).
| Characteristic | |
| Median age, years (range) | 62 (21-91) |
| Male, n (%) | 348 (57.0%) |
| Histologic subtypes, n (%) | |
| Aggressive | 132 (21.6%) |
| Follicular lymphoma-grade III | 19 (3.1%) |
| Mantle cell lymphoma | 30 (4.9%) |
| Diffuse large B-cell lymphoma | 78 (12.8%) |
| Mediastinal/thymic large B-cell lymphoma | 3 (0.5%) |
| Burkitt lymphoma | 2 (0.3%) |
| Indolent | 479 (78.4%) |
| Chronic Lymphocytic Leukemia/Small Lymphocytic Lymphoma | 254 (41.6%) |
| Splenic marginal zone lymphoma | 11 (1.8%) |
| Extranodal marginal zone B-cell lymphoma | 42 (6.9%) |
| Nodal marginal zone B-cell lymphoma | 6 (1.0%) |
| Follicular lymphoma-grade I | 106 (17.4%) |
| Follicular lymphoma-grade II | 58 (9.5%) |
| Other B-cell | 2 (0.3%) |
| ECOG performance status, n (%) | |
| 0 | 432 (70.7%) |
| 1 | 133 (21.8%) |
| 2 | 24 (3.9%) |
| 3+ | 11 (1.8%) |
| Missing | 11 (1.8%) |
| Quality of life, mean (standard deviation) | |
| Physical well-being | 23.9 (4.7) |
| Social/family well-being | 23.5 (4.5) |
| Emotional well-being | 18.0 (4.1) |
| Functional well-being | 20.9 (5.8) |
| FACT-G total | 86.6 (14.2) |
| Follow-up quality of life data, n (% of baseline) | |
| 12 month | 434 (71.0%) |
| 24 month | 422 (69.1%) |
| 36 month | 231 (37.8%) |
Open in a separate window
We closely replicated the analyses conducted by Cella et al. (1993) [5] in the initial validation study for the FACT-G. The validity measures assessed by Cella et al. included convergent and divergent validity, known-groups differentiation, and sensitivity to change. Reliability was assessed with test-retest reliability and Cronbach's alpha, a measure of internal consistency reliability. We replicated the analyses conducted in the initial validation study to the greatest extent possible in our longitudinal sample of NHL patients.
Reliability and item-level descriptive statistics
Cronbach's alpha was calculated for each subscale and for the FACT-G total scale measured at each assessment. Common thresholds for acceptable reliability are >0.70 for group-level applications and >0.90 for individual-level applications [10]. In additional analyses not conducted by Cella et al. [5], we evaluated several item-level psychometrics: item-total correlations (corrected for overlap), percent missing, and floor and ceiling effects. Poor corrected item-total correlation (<0.4) [11] between an item and its respective subscale and high proportion of missing responses (>5%) may be indications that respondents find an item irrelevant. Floor and ceiling effects occur when a large proportion of the respondents have the minimum or maximum possible score, respectively. Items with floor or ceiling effect may be less informative because they lack variability. Items were reverse scored where applicable such that a floor effect indicates poor quality of life and a ceiling effect represents good quality of life. These analyses were conducted separately by phase of treatment (Observation, Active treatment/Progression, or Remission), which was determined at each assessment, to evaluate whether certain questions perform differently by subgroup.
Validity
Convergent validity was assessed by calculating correlations between measures of the same or related constructs. High correlations are evidence of convergent validity [12]. Cella et al. [5] correlated the FACT-G total score with the Functional Living Index – Cancer (FLIC), Brief Profile of Mood States (B-POMS), the Taylor Manifest Anxiety Scale, and patient-reported ECOG Performance Status (ECOG PS). Comparable measures used in the Lymphoma SPORE MER were ECOG PS (measured at all assessments), the B-POMS, and the “state” component of the State-Trait Anxiety Inventory (STAI) (both measured at 36 months). These three variables were correlated with the FACT-G total score as our primary means of assessing convergent validity using the 36-month data, as that was the only assessment in which all of these measures were administered together. We also correlated the FACT-G total score with the linear analog scale assessment (LASA) measure of overall QoL [13] in a secondary analysis to determine the correspondence of the multi-item FACT-G with a single-item summary measure of QoL.
Divergent (or discriminant) validity was assessed by calculating correlations between measures of different constructs, with low correlations being evidence of divergent validity. The initial FACT-G validation used the Marlowe-Crowne Social Desirability Scale for this purpose. As neither this nor a similar scale was included in any of the survey instruments in the Lymphoma SPORE MER, we were unable to estimate divergent validity in this study.
The ability of the FACT-G subscale and total scores to differentiate clinically-distinct groups was assessed using t-tests and ANOVA. Cella et al. [5] assessed the ability of the four subscales and the total scale to differentiate ECOG PS (0, 1, 2, 3/4), stage of illness (I, II, III, IV) and patient location, which was a surrogate for the type of care received. We assessed known-groups validity using ECOG PS at all assessments and disease severity, defined for NHL patients as aggressive vs. indolent disease at the baseline assessment only.
Sensitivity to change
We assessed whether FACT-G subscale and total scale scores changed as expected in patient groups defined by change in a criterion variable. As with the initial validation study [5], we used ECOG PS measured at each assessment as the criterion variable. Patients were classified as “declined” (ECOG PS score increased by at least 1 point), “no change” (change score=0) or “improved” (score decreased by at least 1 point). Univariate ANOVA was used to assess whether FACT-G subscale and total scale change scores can distinguish the three ECOG PS change groups. In a secondary analysis, we used the LASA overall QoL measure to assess sensitivity to change as described above. Patients were categorized as “declined” if their scores decrease by a clinically meaningful amount, defined here as a difference of 2 or more points on an 11-point LASA scale from baseline to a later assessment, [14] “no change” if their scores decreased or increased by less than 2 points, and “improved” if their scores increased by 2 or more points.
RESULTS
Statistical Analysis
Analytic sample
Characteristics at the baseline assessment of the sample selected from the Lymphoma SPORE MER for this validation study are summarized in Table I.
Reliability and item-level descriptive statistics
The FACT-G subscales and total scale for the whole sample combined demonstrated good internal consistency reliability (≥0.7) at all time points, and it was excellent (≥0.9) at all assessments for the FACT-G total scale and at two assessments for the FWB subscale (Figure 1). The reliability for the EWB scale declined over time, but did not go below 0.7. When evaluated by subgroup (Observation, Active treatment/Progression, and Remission), reliability dropped below 0.7 for the EWB scale at one time point in all subgroups, and it was below 0.7 for the SWB scale for the Active treatment/Progression group at only one time point. The items that could be removed from the subscales to increase the reliability to above 0.7 are summarized in Table II.
Figure 1
Internal consistency reliability of FACT-G subscales and total scale over time.
Table II
Summary of reliability and item-level descriptive statistics by subgroup.
| Corrected item-total correlation <0.40 | Percent missing >5% | Deleting item increases its subscale alpha from <0.7 to ≥0.7 | |||||||
|---|---|---|---|---|---|---|---|---|---|
| 12 Mo | 24 Mo | 36 Mo | 12 Mo | 24 Mo | 36 Mo | 12 Mo | 24 Mo | 36 Mo | |
| Active Treatment/Progression | |||||||||
| GP2-nausea | X | ||||||||
| GP3-family needs | X | X | |||||||
| GF2-work fulfilling | X | ||||||||
| GS6-close to partner | X | ||||||||
| GS7-satisfied sex life | X | X | X | X | X | ||||
| GE2-satisf coping | X | X | X | ||||||
| GE4-nervous | X | ||||||||
| GE5-worry dying | X | ||||||||
| Remission | |||||||||
| GP2-nausea | X | X | X | ||||||
| GP4-pain | X | ||||||||
| GP5-side effects | X | ||||||||
| GS7-satisfied sex life | X | X | X | X | X | ||||
| GF4-accept illness | X | ||||||||
| GE2-satisfied coping | X | X | X | X | |||||
| GE3-losing hope | X | X | X | ||||||
| Observation | |||||||||
| GP5-side effects | X | X | X | ||||||
| GS7-satisfied sex life | X | X | X | X | |||||
| GE2-satisfied coping | X | X | X | X | |||||
Open in a separate window
Twelve items did not meet the criterion for corrected item-total correlation of ≥0.4 for at least one subgroup and time point, with more instances occurring in the Remission subgroup (Table II). Item GE2 (satisfied coping with illness) was problematic at multiple time points in all three subgroups. Within the Remission subgroup, three items (GP2: nausea, GE2: satisfied coping with illness, GE3: losing hope in fight against illness) did not meet this criterion at all time points.
No strong patterns of missing data were observed except for item GS7 (satisfied with sex life), which had a very high percent missing (>25%) at all time points in all subgroups. More items were flagged for low item-total correlation or high percent missing at the 36-month assessment than at the 12- and 24-month assessments.
Floor effects were extremely rare. With only a few exceptions, the proportion of responses at the floor was less than 5% for all items in each of the three subgroups at all assessments. Thus, we focus herein on ceiling effects, which were much more prevalent. Ceiling effects were most apparent for the following items in all subgroups at all assessments (Table III): GP2-nausea, GP6-feel ill, GP7-forced to spend time in bed, and GE3-losing hope in fight against illness. In addition, a high proportion of patients in both the Remission and Observation subgroups endorsed the highest response for GP3-meeting needs of family. GP-5 bothered by side effects, had very high ceiling effects (92%-97%) in the Observation group. Since GP5 was reverse scored so that a high score indicated good quality of life, a ceiling effect for this item means that 92%-97% of respondents were not bothered by side effects at all. Overall, the Observation group had more instances of ceiling effects than the Remission group, which had more than the Active treatment/Progression group. At the subscale score level, the EWB subscale had the smallest percent of participants scoring the maximum possible score (range 7.1%-19.4%), and these values were similar to the normative data of 12.6% at ceiling for cancer patients [15]. The FWB subscale had percent with the maximum score ranging from 8.5%-22.9%, with most values higher than the cancer population norm of 9.5%. All values for percent at ceiling were higher than the cancer norm of 11.8% for the PWB subscale (range 14.3%-29.2%) [15], especially in the Observation group where values ranged from 24.2%-29.2%. The percent of participants scoring at the maximum for the SWB subscale was high for all groups ranging from 16.7%-26.4%; the normative percent at ceiling for the SWB subscale is 15.6% [15].
Table III
Proportion of respondents at ceiling (items reversed where applicable such that ceiling represents the best quality of life for that item).
| Item/subscale | Active Treatment/Progression | Remission | Observation | ||||||
|---|---|---|---|---|---|---|---|---|---|
| 12 Mo | 24 Mo | 36 Mo | 12 Mo | 24 Mo | 36 Mo | 12 Mo | 24 Mo | 36 Mo | |
| GP1-lack of energy | 23.4 | 21.6 | 25.6 | 19.5 | 23.6 | 32.3 | 39.1 | 31.5 | 34.2 |
| GP2-nausea | 80.9 | 80.8 | 74.4 | 85.6 | 90.1 | 93.9 | 88.9 | 87.1 | 90.8 |
| GP3-family needs | 59.6 | 67.6 | 57.5 | 74.0 | 78.5 | 84.4 | 80.6 | 79.5 | 80.7 |
| GP4-pain | 78.7 | 69.9 | 73.2 | 56.6 | 65.3 | 66.2 | 70.2 | 66.2 | 63.6 |
| GP5-side effects | 55.3 | 67.6 | 78.1 | 64.7 | 70.4 | 73.4 | 91.8 | 92.8 | 97.1 |
| GP6-feel ill | 72.3 | 77.0 | 78.6 | 78.4 | 88.2 | 87.7 | 84.9 | 82.7 | 85.0 |
| GP7-spend time in bed | 85.1 | 82.4 | 83.7 | 84.4 | 89.6 | 85.9 | 89.8 | 90.0 | 89.2 |
| PWB subscale | 14.9 | 17.6 | 23.3 | 14.3 | 18.9 | 26.2 | 26.9 | 29.2 | 24.2 |
| GE1-feel sad | 30.4 | 54.2 | 45.2 | 49.0 | 53.2 | 49.2 | 46.3 | 49.3 | 55.5 |
| GE2-satisfied coping | 40.4 | 52.7 | 40.9 | 49.4 | 51.1 | 45.5 | 48.3 | 53.5 | 49.2 |
| GE3-losing hope | 74.5 | 79.7 | 80.0 | 86.4 | 91.6 | 93.9 | 88.3 | 84.6 | 88.2 |
| GE4-accepted illness | 52.2 | 55.6 | 51.2 | 51.3 | 50.4 | 52.5 | 52.6 | 53.2 | 60.8 |
| GE5-worry dying | 57.5 | 58.3 | 55.0 | 51.3 | 53.2 | 58.5 | 57.4 | 58.2 | 51.7 |
| GE6-worry get worse | 21.3 | 20.6 | 19.1 | 24.8 | 34.3 | 21.5 | 20.4 | 23.4 | 25.0 |
| EWB subscale | 8.5 | 12.5 | 7.1 | 12.3 | 19.4 | 10.8 | 9.5 | 10.0 | 12.5 |
| GS1-close to friends | 53.2 | 50.0 | 42.9 | 49.7 | 52.1 | 52.5 | 50.7 | 51.5 | 50.0 |
| GS2-support from family | 74.5 | 70.3 | 59.5 | 68.2 | 70.1 | 66.7 | 67.0 | 62.9 | 58.8 |
| GS3-support from friends | 61.7 | 58.1 | 52.4 | 53.9 | 58.0 | 56.9 | 54.9 | 51.8 | 45.3 |
| GS4-family accepted illness | 73.9 | 68.9 | 69.1 | 60.4 | 71.3 | 67.2 | 68.6 | 62.1 | 61.5 |
| GS5-satisfied communication | 63.8 | 64.9 | 61.9 | 64.1 | 71.3 | 66.2 | 66.5 | 62.6 | 60.2 |
| GS6-close to partner | 79.6 | 74.0 | 70.7 | 76.7 | 77.4 | 73.8 | 78.0 | 71.9 | 69.8 |
| GS7-satisfied sex life | 19.4 | 29.4 | 32.0 | 29.0 | 34.3 | 21.7 | 33.9 | 32.3 | 18.8 |
| SWB subscale | 19.2 | 24.3 | 16.7 | 24.7 | 23.6 | 23.1 | 21.6 | 26.4 | 18.3 |
| GF1-able to work | 43.5 | 43.8 | 47.6 | 49.0 | 52.4 | 56.1 | 62.9 | 63.5 | 62.5 |
| GF2-work fulfilling | 34.1 | 41.7 | 46.3 | 39.0 | 45.5 | 40.9 | 47.0 | 48.5 | 49.2 |
| GF3-able to enjoy life | 42.6 | 52.7 | 52.4 | 46.8 | 57.3 | 57.6 | 60.6 | 58.7 | 57.5 |
| GF4-accepted illness | 53.2 | 59.5 | 59.5 | 53.3 | 59.0 | 45.5 | 55.8 | 54.7 | 60.0 |
| GF5-sleeping well | 29.8 | 36.5 | 35.7 | 30.1 | 32.6 | 35.4 | 37.2 | 32.5 | 38.3 |
| GF6-enjoy things for fun | 44.7 | 46.6 | 45.2 | 37.7 | 44.4 | 43.9 | 51.1 | 49.0 | 50.0 |
| GF7-content quality of life | 40.4 | 48.7 | 35.7 | 37.7 | 49.3 | 40.9 | 52.0 | 46.3 | 50.4 |
| FWB subscale | 8.5 | 13.5 | 14.3 | 13.0 | 16.7 | 15.2 | 22.9 | 18.4 | 17.5 |
| FACT-G Total | 0.0 | 1.4 | 2.4 | 0.5 | 5.6 | 3.1 | 2.2 | 3.5 | 0.8 |
Open in a separate window
Validity
Correlations at 36 months between the FACT-G total score and ECOG PS (r = -0.43, n = 177), B-POMS (r = -0.63, n = 201), STAI (r = -0.57, n = 226) and LASA QoL (r = 0.63, n = 229) were moderate to large in magnitude, were in the expected direction and were statistically significant (all p < 0.001).
Although Cella et al. compared FACT-G scores across four categories of ECOG PS (0, 1, 2, 3/4), there were too few patients with ECOG PS greater than 2 in our sample at any time point; thus, categories 2, 3, and 4 were combined for this analysis. However, sample sizes for the combined 2+ category remained small (range 13-35). The FACT-G total and subscale scores were able to differentiate patients based on their ECOG PS at all time points with only two exceptions: SWB and EWB scores did not differentiate patients at the 36-month assessment (Table IV). All pair-wise comparisons were consistently statistically significant for the FACT-G total, PWB and FWB scores and the mean differences across groups exceeded the minimally important differences of 5-7 points for the FACT-G total [16] and 2-3 points for the PWB and FWB [17]. Mean scores between patients with indolent and aggressive disease at baseline were statistically significant (p < 0.001) for the FACT-G total (score difference=6.4 points), PWB (2.7 points) and FWB (3.5 points). EWB and SWB scores did not differ significantly by disease severity.
Table IV
FACT-G differentiation of ECOG Performance Status.
| Mean FACT-G scores | |||||
|---|---|---|---|---|---|
| Scale | Assessment | PS 0 | PS 1 | PS 2+ | F-test p-value & significant pair-wise comparisons |
| FACT-G total | Baseline n=588 | 90.5 | 78.9 | 65.7 | <0.001 0>1>2+ |
| 12 Months n=427 | 94.6 | 82.6 | 62.2 | <0.001 0>1>2+ | |
| 24 Months n=415 | 94.9 | 85.6 | 68.0 | <0.001 0>1>2+ | |
| 36 Months n=177 | 93.7 | 86.0 | 75.5 | <0.001 0>1>2+ | |
| PWB | Baseline n=597 | 25.5 | 21.0 | 16.1 | <0.001 0>1>2+ |
| 12 Months n=428 | 26.3 | 22.5 | 15.2 | <0.001 0>1>2+ | |
| 24 Months n=421 | 26.4 | 22.7 | 17.7 | <0.001 0>1>2+ | |
| 36 Months n=177 | 26.5 | 23.4 | 19.3 | <0.001 0>1>2+ | |
| SWB | Baseline n=596 | 23.9 | 22.8 | 21.5 | 0.002 0>1,2+ |
| 12 Months n=428 | 24.0 | 22.4 | 21.3 | 0.001 0>1 | |
| 24 Months n=420 | 24.1 | 22.7 | 21.1 | <0.001 0>1,2+ | |
| 36 Months n=177 | 22.7 | 23.2 | 21.4 | NS | |
| EWB | Baseline n=593 | 18.3 | 17.3 | 16.7 | 0.01 0>1 |
| 12 Months n=431 | 20.0 | 18.7 | 14.6 | <0.001 0>1>2+ | |
| 24 Months n=417 | 20.1 | 19.3 | 16.4 | 0.001 0,1>2+ | |
| 36 Months n=177 | 20.2 | 19.2 | 19.3 | NS | |
| FWB | Baseline n=595 | 22.7 | 17.8 | 11.7 | <0.001 0>1>2+ |
| 12 Months n=431 | 24.3 | 19.1 | 11.8 | <0.001 0>1>2+ | |
| 24 Months n=420 | 24.3 | 20.4 | 12.8 | <0.001 0>1>2+ | |
| 36 Months n=178 | 24.4 | 20.2 | 15.5 | <0.001 0>1>2+ | |
Open in a separate window
*Scheffe pairwise comparisons; > symbol separates groups that report significantly (p < 0.05) higher scores from those with lower scores. PS: performance status, PWB: physical well-being, SWB: social/family well-being, EWB, emotional well-being, FWB, functional well-being, NS: not statistically significant
Sensitivity to change
The FACT-G total score and subscale scores demonstrated very good sensitivity to change in ECOG PS and LASA QoL with a few exceptions (Tables V and andVI).VI). The SWB subscale was not sensitive to change in ECOG PS for any of the three change periods and was sensitive to change in LASA QoL for the baseline to 12-month change period only. The EWB subscale was not sensitive to change in ECOG PS for the baseline to 12-month and baseline to 36-month change period.
Table V
FACT-G sensitivity to change in ECOG Performance Status.
| Mean FACT-G change scores | |||||
|---|---|---|---|---|---|
| Scale | Change Period | Declined PS | No change in PS | Improved PS | F-test p-value & significant pairwise comparisons |
| FACT-G total | BL to 12 Mo n=410 | -2.0 | 2.2 | 9.7 | <0.001 D<N<I |
| BL to 24 Mo n=401 | -3.6 | 2.6 | 14.7 | <0.001 D<N<I | |
| BL to 36 Mo n=171 | -3.9 | 4.0 | 9.9 | <0.001 D<N,I | |
| PWB | BL to 12 Mo n=415 | -1.1 | 0.3 | 4.0 | <0.001 D<N<I |
| BL to 24 Mo n=408 | -1.2 | 0.4 | 4.6 | <0.001 D<N<I | |
| BL to 36 Mo n=223 | -1.7 | 0.2 | 3.8 | <0.001 D<N<I | |
| SWB | BL to 12 Mo n=417 | -0.8 | -0.1 | -0.7 | NS |
| BL to 24 Mo n=410 | -1.3 | -0.3 | 0.0 | NS | |
| BL to 36 Mo n=174 | 0.0 | -0.2 | -1.0 | NS | |
| EWB | BL to 12 Mo n=418 | 0.9 | 1.3 | 1.9 | NS |
| BL to 24 Mo n=407 | 0.2 | 1.7 | 3.3 | <0.001 D<N<I | |
| BL to 36 Mo n=173 | 0.9 | 2.4 | 1.7 | NS | |
| FWB | BL to 12 Mo n=419 | -0.9 | 0.9 | 4.3 | <0.001 D<N<I |
| BL to 24 Mo n=411 | -1.3 | 0.8 | 6.3 | <0.001 D<N<I | |
| BL to 36 Mo n=174 | -1.6 | 1.7 | 4.7 | <0.001 D<N,I | |
Open in a separate window
*Scheffe pairwise comparisons; < symbol separates groups that report significantly (p < 0.05) lower change scores from those with higher change scores. PS: performance status, PWB: physical well-being, SWB: social/family well-being, EWB, emotional well-being, FWB, functional well-being, BL: Baseline, Mo: Months, D: declined, N: no change, I: improved, NS: not statistically significant
Table VI
FACT-G sensitivity to change in overall Quality of Life Linear Analog Self-Assessment.
| Mean FACT-G change scores | |||||
|---|---|---|---|---|---|
| Scale | Change period | Declined QoL LASA | No change in QoL LASA | Improved QoL LASA | F-test p-value & significant pairwise comparisons |
| FACT-G total | BL to 12 Mo n=410 | -9.4 | 2.2 | 11.3 | <0.001 D<N<I |
| BL to 24 Mo n=401 | -5.4 | 1.5 | 11.4 | <0.001 D<N<I | |
| BL to 36 Mo n=171 | -8.5 | 1.6 | 11.7 | <0.001 D<N<I | |
| PWB | BL to 12 Mo n=415 | -3.0 | 0.3 | 3.9 | <0.001 D<N<I |
| BL to 24 Mo n=408 | -1.8 | 0.1 | 3.6 | <0.001 D<N<I | |
| BL to 36 Mo n=223 | -3.1 | -0.1 | 3.9 | <0.001 D<N<I | |
| SWB | BL to 12 Mo n=417 | -3.0 | -0.1 | 0.7 | <0.001 D<N,I |
| BL to 24 Mo n=410 | -1.9 | -0.3 | 0.1 | NS | |
| BL to 36 Mo n=174 | -0.6 | -0.8 | -0.3 | NS | |
| EWB | BL to 12 Mo n=418 | -0.2 | 1.3 | 2.2 | 0.002 D<N,I |
| BL to 24 Mo n=407 | 0.2 | 1.3 | 3.6 | <0.001 D,N<I | |
| BL to 36 Mo n=173 | -0.4 | 1.8 | 4.0 | <0.001 D<N<I | |
| FWB | BL to 12 Mo n=419 | -3.3 | 0.7 | 4.7 | <0.001 D<N<I |
| BL to 24 Mo n=411 | -2.6 | 0.5 | 4.6 | <0.001 D<N<I | |
| BL to 36 Mo n=174 | -4.3 | 0.8 | 4.0 | <0.001 D<N<I | |
Open in a separate window
*Scheffe pairwise comparisons; < symbol separates groups that report significantly (p < 0.05) lower change scores from those with higher change scores. LASA: linear analog self-assessment, PWB: physical well-being, SWB: social/family well-being, EWB, emotional well-being, FWB, functional well-being, BL: Baseline, Mo: Months, D: declined, N: no change, I: improved, NS not statistically significant
DISCUSSION
Reliability and Item-Level Descriptive Statistics
We evaluated item-level statistics to identify specific FACT-G questions that may be problematic when monitoring QoL over time in NHL patients. The item with poor psychometric properties in all subgroups at multiple assessments was GS7 “I am satisfied with my sex life.” However, due to the sensitive nature of this question, it is often problematic [8,18,19], and neither the poor item-total correlation nor high percent missing is unique to NHL patients. The prorating in the scoring algorithm accounts for the high percent missing for this item. Another question that was problematic across subgroups was GE2 “I am satisfied with how I am coping with my illness.” This finding was somewhat unexpected and could not be explained by ceiling effects. To better interpret this finding, we reviewed the raw data for the EWB subscale and observed that in a small number of observations, it appeared as though patients “straight-lined” their responses. That is, they selected one response category for all six questions in the subscale not recognizing that the question wording for GE2 was positive, and therefore, different from the other five negatively worded questions (e.g. GE5, “I worry about dying”). The EWB subscale is the only FACT-G subscale in which the direction of items switches within the subscale. This type of response bias (i.e. position bias) may occur if a respondent is careless or fatigued when completing the questionnaire, and it is a function of questionnaire length [20]. Although the FACT-G is not a particularly long questionnaire, it was administered along with other measures as part of larger survey packets. For example, the 2-page FACT-G questionnaire was part of the 32-page baseline survey packet.
A potential straight-line effect is most obvious when one extreme of the response scale (i.e. either 0 or 4) was selected for all questions. We conducted an unplanned sensitivity analysis to determine whether excluding data for participants who straight-lined one of the extreme responses (either 0 or 4) for all six EWB questions changed the interpretation of the results. Extreme straight-lining was observed in 17 (2.8%) subjects at baseline, 14 (3.2%) at 12 months, 19 (4.5%) at 24 months and 10 at 36 months (4.4%). GE2 had 11 instances of poor item-level psychometrics in Table II. Four of the 11 instances were resolved when the straight-line data were excluded. Of the remaining instances, the psychometrics tended to improve even if they did not exceed the established threshold. For example, although the corrected item-total correlation for GE2 in the remission subgroup at the 24-month assessment increased substantially from 0.18 to 0.35 when the straight-line data were removed, it remained below the criterion of 0.4. Mean EWB scores and subscale-level psychometrics (e.g. known-groups validity) improved slightly when the straight-line data were removed, but the overall conclusions were unchanged.
This observation led us to consider straight-line effects across subscales. All items in the PWB and EWB subscales are negatively worded (except for GE2 as previously discussed); whereas all items in the FWB and SWB subscales are positively worded. The PWB and SWB appear together on one page of the questionnaire and the EWB and FWB appear on another. We identified cases with extreme discrepancies in subscale scores (e.g. highest possible score on the PWB subscale and lowest possible score on the SWB subscale). In a second, ad hoc sensitivity analysis, we determined whether deleting these cases affected the interpretation of item-level or subscale-level psychometrics. Instances of straight-lining across scales on the same page of the questionnaire was much less common and occurred among 3 participants at baseline, 2 at 12 months, 4 at 24 months, and 3 at 36 months. Excluding these cases did not affect the item-level or subscale-level results or the overall conclusions (data not shown).
Ceiling effects were observed at the item level and the subscale level, and the percent at ceiling tended to be highest for the Observation group, followed by the Remission and Active Treatment/Progression groups. Ceiling effects often occur when an existing instrument is used in a population that differs from the one in which it was originally developed [21]. Of the groups studied here, the Active Treatment/Progression group is the most consistent with the intended target patient population for the FACT instruments. Thus, ceiling effects would be expected to be more common in the Remission and Observation groups. Ceiling effects are of concern for longitudinal analyses since the scale is unable to measure improvement for individuals already at the maximum score [21].
Validity
Our results were very comparable to those reported in Cella et al. The magnitude of the correlations for ECOG PS and B-POMS were similar to those reported by Cella et al. (STAI and LASA QoL were not measured in Cella et al.). Evidence of known-groups validity was weaker for the EWB and SWB subscales, as indicated by the lack of significance in the omnibus F-tests at 36 months and for several pair-wise comparisons at other assessments. The ability of the SWB subscale to differentiate ECOG PS was also weaker than that of other scales in Cella et al. However, these results may be more of a reflection on the use of ECOG PS as the criterion variable rather than an indictment of the validity of the SWB subscale in NHL patients. It is possible that SWB is resilient to physical impairment and resulting limitations as measured by ECOG PS. The SWB subscale primarily measures emotional social support (as opposed to instrumental social support) with items such as “I get emotional support from my family” and “I get support from my friends.” Arora et al. found that emotional support from family and friends was consistently perceived as helpful in a longitudinal study of breast cancer patients, and that a correlate of perceived helpfulness was recent cancer treatment [22]. Thus, it is possible that when patients experience physical limitations, as would be expected following treatment for cancer, they may request more support from friends and family, or they may be more likely to recognize the existing support provided by friends and family to help them manage those limitations.
Sensitivity to Change
The FACT-G total scale, PWB and FWB consistently showed very good sensitivity to change in ECOG PS, whereas the EWB and SWB subscales did not. As with the known-groups validity, these results may be due to dissimilarity in constructs measured by ECOG PS and the EWB or SWB subscales. The inability to detect change may also be due in part to the ceiling effects observed. SWB change scores did not map onto LASA QoL change scores for two of the three change periods evaluated. To help interpret this finding, we computed correlations between the subscales and LASA QoL at each of the four assessments (data not shown). Correlations with LASA QoL were highest at all assessments for FWB, followed by PWB then EWB. SWB consistently had the smallest correlation with LASA QoL. Thus, it appears as though patients reflect more on physical than psychosocial aspects of their health when answering a global QoL question, a finding previously reported by others [13,23].
Limitations
Ideal criterion variables for assessing the validity and sensitivity to change of the EWB and SWB subscales were not available. Additional analyses to confirm the validity and sensitivity to change of these subscales in patients with NHL are needed. Possible criterion variables include a measure of social support such as the MOS Social Support scale [24] to compare to the SWB. The STAI and B-POMS are suitable criterion variables for the EWB, but were only assessed at one time point. Longitudinal data for the STAI, B-POMS, MOS Social Support scale or other suitable criterion variables would allow for the assessment of sensitivity to change in the EWB and SWB subscales.
We replicated the validation analyses reported by Cella et al. [5] and we demonstrated that the FACT-G had acceptable psychometric properties in a sample of 611 NHL patients. We also demonstrated that the FACT-G can be used to monitor QoL over time in this population.
The FACT-G contains several items referencing illness or treatment that may be less relevant to a substantial proportion of patients at the later assessments (i.e. 24, 36 months after diagnosis) who are in observation or remission. These items include “I have nausea” (GP2); “I am bothered by side effects of treatment” (GP5); and “I am losing hope in the fight against my illness” (GE3). While the item “I am satisfied with how I am coping with my illness” (GE2) was problematic across subgroups, this may be due to the reverse wording within the subscale rather than a lack of relevance to NHL patients. Steps can be taken within a study protocol to minimize this error, such as reviewing each questionnaire for straight-lining within the EWB subscale and following-up with patients to confirm suspicious response. Changing GE2 to be a negatively worded item (e.g. “I am dissatisfied with how I am coping with my illness”) may also remedy this issue, but would require approval from the instrument developers and further validation work, including cognitive testing among patients with cancer.
The selection of an instrument to measure QoL in NHL should consider balancing relevance and specificity and should also consider the target patient population. A more general instrument is likely to be relevant to broad range of patients, and may be better suited for a study comprising patients only in remission or on observation. An NHL-specific instrument is likely to detect disease- or treatment-related concerns and may be better suited for a study comprising patients only in active treatment/progression. Our results suggest that the FACT-G provides a reasonable balance of relevance and specificity for measuring QoL over time in a patient population consisting of all three groups.
Although our overall conclusion is the that FACT-G is a valid measure of QoL in NHL patients, investigators may consider conducting sensitivity analyses in which the three items listed above are excluded from their respective subscales to determine whether results are resilient to a potential lack of relevance in certain NHL patient subgroups.
ACKNOWLEDGEMENTS
This work was supported by grant P50 CA97274 (University of Iowa/Mayo Clinic Lymphoma SPORE). The authors thank Dr. Jeff Sloan for advising on the selection of quality of life and other patient-reported outcome measures. Portions of this manuscript were presented at the International Society of Quality of Life Research (ISOQOL) annual conference, October 27-30, 2010, London UK.
REFERENCES
1. American Cancer Society . Cancer Facts and Figures 2010. American Cancer Society; Atlanta, GA: 2010. [Google Scholar]
2. Swerdlow SH, Campo E, Harris NL, et al., editors. WHO Classification of Tumours of Haematopoietic and Lymphoid Tissues. IARC; Lyon: 2008. [Google Scholar]
3. The International Non-Hodgkin's Lymphoma Prognostic Factors Project A predictive model for aggressive non-Hodgkin's lymphoma. The International Non-Hodgkin's Lymphoma Prognostic Factors Project. N Engl J Med. 1993;329:987–994. [PubMed] [Google Scholar]
4. Drake MT, Maurer MJ, Link BK, et al. Vitamin D insufficiency and prognosis in non-Hodgkin's lymphoma. J Clin Oncol. 2010;28:4191–4198. [PMC free article] [PubMed] [Google Scholar]
5. Cella D, Tulsky DS, Gray G, et al. The Functional Assessment of Cancer Therapy scale: development and validation of the general measure. J Clin Oncol. 1993;11:570–579. [PubMed] [Google Scholar]
6. Ware J, Snow K, Kosinski M. SF-36 Health Survey: Manual and Interpretation Guide. QualityMetric Incorporated; Lincoln, RI: 2000. [Google Scholar]
7. Webster K, Cashy J, Cella D. Measuring quality of life (QOL) in patients with non-Hodgkin's lymphoma (NHL): The Functional Assessment of Cancer Therapy - Lymphoma (FACT-Lym).. 12th Annual Conference of the International Society for Quality of Life Research; San Francisco, CA: Springer. 2005. [Google Scholar]
8. Fairclough DL, Cella DF. Functional Assessment of Cancer Therapy (FACT-G): non-response to individual questions. Qual Life Res. 1996;5:321–329. [PubMed] [Google Scholar]
9. Cella D, editor. Manual of the Functional Assessment of Chronic Illness Therapy (FACIT) Measurement System, Version 4.1. Center on Outcomes, Research & Education (CORE); Evanston, IL: 2004. [Google Scholar]
10. Hahn EA, Cella D, Chassany O, Fairclough DL, Wong GY, Hays RD. Precision of health-related quality-of-life data compared with other clinical measures. Mayo Clin Proc. 2007;82:1244–1254. [PubMed] [Google Scholar]
11. McHorney CA, Ware JE, Lu JFR, Sherbourne CD. The MOS 36-Item Short-Form Health Survey (SF-36): III. Tests of data quality, scaling assumptions, and reliability across diverse patient groups. Med Care. 1994;32:40–66. [PubMed] [Google Scholar]
12. Crocker L, Algina J. Introduction to Classical & Modern Test Theory. Harcourt Brace Jovanovich College Publishers; Orlando, FL: 1986. [Google Scholar]
13. Locke DE, Decker PA, Sloan JA, et al. Validation of single-item linear analog scale assessment of quality of life in neuro-oncology patients. J Pain Symptom Manage. 2007;34:628–638. [PMC free article] [PubMed] [Google Scholar]
14. Farrar JT, Young JP, Jr., LaMoreaux L, Werth JL, Poole RM. Clinical importance of changes in chronic pain intensity measured on an 11-point numerical pain rating scale. Pain. 2001;94:149–158. [PubMed] [Google Scholar]
15. Brucker PS, Yost K, Cashy J, Webster K, Cella D. General population and cancer patient norms for the Functional Assessment of Cancer Therapy-General (FACT-G). Eval Health Prof. 2005;28:192–211. [PubMed] [Google Scholar]
16. Cella D, Hahn EA, Dineen K. Meaningful change in cancer-specific quality of life scores: differences between improvement and worsening. Qual Life Res. 2002;11:207–221. [PubMed] [Google Scholar]
17. Yost KJ, Eton DT. Combining distribution- and anchor-based approaches to determine minimally important differences: The FACIT experience. Eval Health Prof. 2005;28:172–191. [PubMed] [Google Scholar]
18. Ashing-Giwa KT, Kim J, Tejero JS. Measuring quality of life among cervical cancer survivors: preliminary assessment of instrumentation validity in a cross-cultural study. Qual Life Res. 2008;17:147–157. [PubMed] [Google Scholar]
19. Smith AB, Wright P, Selby PJ, Velikova G. A Rasch and factor analysis of the Functional Assessment of Cancer Therapy-General (FACT-G). Health Qual Life Outcomes. 2007;5:19. [PMC free article] [PubMed] [Google Scholar]
20. Herzog A, Bachman J. Effects of questionnaire length on response quality. Public Opin Q. 1981;45:549–559. [Google Scholar]
21. de Vet HC, Terwee CB, Mokkink L, Knol D. Measurement in Medicine: A Practical Guide. Cambridge University Press; New York: 2011. [Google Scholar]
22. Arora NK, Finney Rutten LJ, Gustafson DH, Moser R, Hawkins RP. Perceived helpfulness and impact of social support provided by family, friends, and health care providers to women newly diagnosed with breast cancer. Psychooncology. 2007;16:474–486. [PubMed] [Google Scholar]
23. Beijer S, Kempen GI, Pijls-Johannesma MC, de Graeff A, Dagnelie PC. Determinants of overall quality of life in preterminal cancer patients. Int J Cancer. 2008;123:232–235. [PubMed] [Google Scholar]
24. Sherbourne CD, Stewart AL. The MOS social support survey. Soc Sci Med. 1991;32:705–714. [PubMed] [Google Scholar]
