Introduction

Sarcopenia is a progressive loss of muscle mass and function that is related to ageing.1 It leads to functional impairment,2, 3 increased risk of falls and fractures,4-7 loss of independence,8, 9 and an increased overall mortality.10-13 Approximately 5–13% of adults aged 60–70 years are affected, increasing to 11–50% for those aged ≥80 years.14 Thus, it constitutes a major public health issue.15-18 Several adverse outcomes are associated with sarcopenia, such as an increased rate of post-operative complications after gastrointestinal surgery,19 and a worse recovery during rehabilitation20 and strokes.20, 21 Therefore, the diagnosis and implementation of sarcopenia into treatment plans gain more interest in a broad variety of medical specialties.

Several different definitions of sarcopenia have been propagated. We refer to the revised definition of the European Working Group on Sarcopenia in Older People,1 where a diagnosis of probable sarcopenia is given in case of reduced muscle strength and further confirmed by the finding of reduced muscle quantity or quality.1 To assess the severity of sarcopenia, testing physical performance is advised, for example, with the ‘short physical performance battery’ (SPPB).1 SPPB test scores of <9 indicate a high risk for severe sarcopenia-related functional impairments.1 For sarcopenia screening, the revised definition of the European Working Group on Sarcopenia in Older People recommends the use of the strength, assistance walking, rise from a chair, climb stairs, and falls (SARC-F) questionnaire.1

The SARC-F questionnaire contains five self-reported items evaluating the hallmarks of sarcopenia, that is, functional deficits and falls.22 Each item has a possible score of 0 to 2 points, with higher scores suggesting a higher risk of sarcopenia.22 A total score of >3 is regarded as a positive screening result.22 Higher SARC-F scores have been shown to correlate with a slower chair rise, lower gait speed, overall lower SPPB scores, as well as adverse outcomes related to sarcopenia.22 The SARC-F has a low-to-moderate sensitivity but high specificity,23, 24 thus providing a well-suited screening test to identify individuals who are not at high risk of sarcopenia-related negative outcomes. Its validation in many different languages eases its use in clinical practice.23, 25-30

So far, the SARC-F questionnaire has only been validated for self-reporting. A person's self-reporting can be inaccurate due to cognitive impairment,31 a non-objective perception of one's own functional capacities,32 or even negation of functional decline. Furthermore, self-reporting is impossible in patients unable to answer questions due to an acute medical condition or altered consciousness.33 As treatment decisions and modifications might benefit from knowledge about the presence of sarcopenia and sarcopenia-related functional impairments as part of a geriatric assessment,34-37 proxy-reported information could be valuable to allow for patient-centred treatment strategies. If the illness has rapidly evolved within a few weeks, information on the patient's premorbid functional impairments might be crucial to differentiate between disease-related and premorbid conditions. The latter could reflect the patient's limited intrinsic physical resilience and guide the physicians towards adapted treatment intensities, such as an adjusted chemotherapy or radiotherapy regimen for the treatment of cancer. In contrast, the misperception of a disease-related deterioration as an intrinsic condition could potentially preclude the patient from a curative approach.

To our knowledge, no screening tool for sarcopenia-related functional impairments has been validated for the use of proxy-reported information so far. To close this gap, we aimed to validate a proxy-reported version of the SARC-F questionnaire for the evaluation of both a patient's current as well as retrospective, premorbid status.

Methods

A proxy-reported version of the original SARC-F questionnaire was developed substituting ‘you’ with the respective patient's name within the questionnaire. For retrospective evaluation, verbs were transformed from present into past tense. No further modifications were made.

Study population

The study population included patients aged ≥60 years and their proxies. Patients undergoing an in-house geriatric rehabilitation for diverse medical conditions at Agaplesion Bethanien Hospital (Heidelberg, Germany; Centre 1) as well as patients under surveillance for a rheumatological or haematological disease at the Heidelberg University Hospital outpatient clinic (Centre 2) were recruited. We selected these diverse centres with the clear intention of recruiting a representative sample of older adults with different levels of functional impairments in the hospital setting. Although patient characteristics within Cohorts A and B were further subdivided into Centres 1 and 2 to provide the detailed composition of these cohorts (e.g. Table 1), patients from Centres 1 and 2 did not represent independent groups and were not statistically separately analysed. Patients were excluded if they were unable to designate at least one close contact person (proxy), suffered from an acute medical condition that precluded them from performing the SPPB or were unable to give informed consent (due to an at least moderate cognitive impairment or an acute alteration of consciousness).

Table 1. General patient and proxy characteristics Characteristics Cohort A Cohort B C1 C2 Total C1 C2 Total Patients Patients (n) 15 49 64 9 31 40 Age [years] (mean ± SD) 81.9 ± 8 (range 63–95) 76.6 ± 5.4 (range 64–90) 77.9 ± 6.5 (range 63–95) 83.6 ± 7.1 (range 72–93) 79.3 ± 5.1 (range 70–91) 80.2 ± 5.8 (range 70–93) Sex (%) Female 73.3 40.8 48.4 55.6 38.7 42.5 SARC-F > 3 points (%) 73.3 26.5 37.5 77.8 16.1 30.0 SPPB < 9 points (%) 93.3 24.5 40.6 88.9 22.6 37.5 Gait speed < 0.8 m/s (%) 80.0 28.6 40.6 88.9 16.1 32.5 Number of participating proxies 1 80% 65.3% 68.8% 88.9% 80.7% 82.5% 2 20% 26.5% 25% 11.1% 19.4% 17.5% 3 0% 8.2% 6.2% 0% 0% 0% Proxies Age [years] (mean ± SD) 66.1 ± 13.4 (range 46–89) 67.3 ± 12.8 (range 34–87) 67 ± 12.8 (range 34–89) 59 ± 16.8 (range 25–82) 65.3 ± 13.4 (range 38–83) 63.9 ± 14.2 (range 25–83) Sex (%) Female 66.7 69.4 68.8 66.7 83.9 80 Relation to patient Partner 26.7 59.2 51.6 22.2 61.3 52.5 Daughter 40.0 10.2 17.2 33.3 25.8 27.5 Son 13.3 18.4 17.2 22.2 9.7 12.5 Siblings 6.7 0 1.6 0 0 0 Brother-/sister-in-law 0 2.0 1.6 0 0 0 Niece/nephew 0 4.1 3.1 0 0 0 Friend 13.3 6.1 7.8 0 3.2 2.5 Grandchild 0 0 0 11.1 0 2.5 Professional caregiver 0 0 0 11.1 0 2.5 Geographical distance to patients Same address 33.3 69.4 60.9 11.1 67.7 55 Different town 66.7 30.6 39.1 55.6 22.6 30 C1, Centre 1 (Agaplesion Bethanien Hospital Heidelberg, geriatric hospital); C2, Centre 2 (University Hospital Heidelberg, rheumatology and haematology outpatient services).

The patient's designated proxies (caregivers, partners, children, siblings, grandchildren, neighbours, and friends) were also invited to participate. They were eligible if they were aged ≥18 years and had been in weekly contact with the patient during the past 6 months. Alternatively, contact on a weekly basis at least via telephone and with also meeting the patient in person at least twice during the past 6 months was accepted. Up to three proxies per patient were included. We classified proxies as main proxy (partner; if no partner was available, the proxy who was claimed as closest contact person by the patient) and additional proxies. We did not stratify patient and proxy recruitment according to gender to avoid any bias that could modify the choice of proxies as the person who knows the patient best. The trial flowchart is depicted in Figure 1.

Trial procedures and recruitment flow chart. Patients performed SPPB and answered SARC-F questionnaire on Day 0 (=T1). Proxies in Cohort A answered SARC-F questionnaire on Day 0 and again retrospectively after 3 months (=T2). Proxies in Cohort B answered the questionnaire only once after 3 months (=T2) to test for a possible recall effect.

Data collection and trial measurements

Patients and their respective proxies were recruited into two different cohorts in chronological order (Cohorts A and B). In both cohorts, the participating patients completed the self-reported SARC-F questionnaire and the SPPB once at baseline (T1). Proxies in Cohort A completed the proxy-reported SARC-F questionnaire twice, first at T1 (simultaneously to the patients' self-report) and then again after 3 months (T2). At T2, the proxies were asked to retrospectively evaluate the patients' functional status at T1. Proxies in Cohort B were not questioned at T1, but called for the first time at T2 to complete the proxy-reported SARC-F only retrospectively (i.e. evaluating the respective patient's functional status at T1). The Cohort B was established to examine a potential recall bias as proxies could potentially recall their first assessment of SARC-F questionnaire at T1 rather than assessing the premorbid condition of the patient when asked to assess the SARC-F retrospectively at T2. The presence of a recall bias would potentially limit the validity of a retrospective assessment. Study procedures are outlined in Figure 1.

All patients and proxies were fluent in German; nonetheless, they were offered to answer the SARC-F questionnaire in their respective mother tongue, which was preferred by three patients (once in Spanish and twice in Turkish).

The SPPB was performed as outlined elsewhere.38 In brief, all patients carried out a balance test, performing a stand in side-by-side-position, semi-tandem, and full-tandem for 10 s each. Thereafter, they were asked to walk a distance of 4 m in their usual gait speed twice; the faster walk was selected for analysis. Walking aids were allowed if used in daily routine. Finally, the patients performed the chair-rise test, getting up from a chair five times as quickly as possible without using the upper extremities or other assistance. The respective results of the three tests were each transformed into a score ranking from 0 to 4 points and summarized to yield an overall SPPB score ranking from 0 to 12 points, with higher scores indicating better performance.

Statistical analyses

Descriptive statistics were reported in absolute numbers and percentages for categorical variables. Possible differences in patient and proxy characteristics between the two study groups were assessed by Fisher's exact and Kruskal–Wallis test. All SARC-F scores were dichotomized into ≤3 (probably not at risk for sarcopenia-related functional impairments) and >3 points (positive screening for severe sarcopenia-related functional impairments). The SPPB score was dichotomized into ≥9 and <9 points.1 The results of the chair-rise test were converted into chair rises per minute (CRPM) from the absolute time needed for five chair rises to allow for a natural numerical representation of a chair-rise test in cases where patients were not able to stand up five times in a row. If a patient was unable to perform this test, the result was set as infinite. Thereby, the introduction of artificial numbers for analysis was avoided.

The SARC-F questionnaire was primarily established to screen for persons who are at risk for sarcopenia-related poor functional outcomes.22 Thus, defining a SPPB score <9 points as the cut-off value for patients at high risk for severe sarcopenia-related functional impairments, we assessed the diagnostic values [sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV), and positive and negative likelihood ratios (PLR, NLR)] of the patient-reported and the different proxy-reported SARC-F screening modalities. We used receiver operating characteristic (ROC) curves to compare the overall accuracy of the patient-reported and proxy-reported SARC-F variants. The areas under the ROC curve (AUCs) were calculated. Comparisons between the AUC of the different SARC-F screening modalities were performed using the DeLong test.39

To assess the agreement between patient-reported and proxy-reported SARC-F results, Cohen's kappa was calculated and rated according to the strength of agreement.40, 41

The results of the patient-reported and the different proxy-reported SARC-F versions were each correlated with the patient's SPPB scores and with their CRPM results. The latter were specifically analysed because SARC-F items mainly focus on lower extremity strength, and may therefore be well reflected by the chair-rise test component of the SPPB.1

The resulting correlation coefficients were then tested against each other in pairwise comparisons for non-inferiority of proxy-reported in comparison with patient-reported SARC-F results to an alpha level of 0.05 with a non-inferiority margin of δ0 = 0.2.42 Additionally, the 95% confidence intervals of the differences between these pairs of correlation coefficients were calculated. In an exploratory analysis, the obtained correlation coefficients were also tested against each other for superiority to an alpha level of 0.05.

Unless stated otherwise, the presented analyses were all performed for the group of the main proxies only. No composite score of different proxies was generated.

To compare the proxy-reported SARC-F results obtained from the different proxy subgroups (main proxies, additional proxies, and proxies with a residential distance), correlations of the retrospective proxy-reported SARC-F results with the CRPM were calculated. The 95% confidence intervals of the differences between the respective correlation coefficients were provided. No formal non-inferiority/superiority testing was performed due to the small sample sizes of the different proxy subgroups.

The sample size for the Cohort A was based on the feasibility to complete recruitment within less than 6 months; however, to ensure proper powering of the study, comprehensive sample size calculations based on different scenarios were conducted. The sample size of the Cohort B was calculated based on the correlation coefficients of the patient-reported SARC-F results with the CRPM of the first 47 patients in Cohort A, aiming at a significance level of ɑ = 0.2 to achieve an empirical power of 88% for non-inferiority.

All analyses were performed using the programming language ‘R’, ‘cocor’ for calculation of correlation coefficients, and ‘pROC’ for generation of ROC curves and AUC calculations.

Results Study population

One hundred and eleven patients and 137 proxies were enrolled in this study; the final study population comprised 104 patients and 135 proxies (Figure 1). Of these, 64 patients and their corresponding proxies (64 main and 24 additional proxies) were included in Cohort A between September 2018 and January 2019. Fifteen of the patients were undergoing geriatric in-patient rehabilitation (Centre 1), and 49 were community-dwelling older adults having a scheduled outpatient visit at the rheumatology or haematology outpatient service (Centre 2). A further 40 patients and their proxies (40 main and 7 additional proxies) were assigned to Cohort B (9 at Centre 1 and 31 at Centre 2) between November 2018 and May 2019. In general, compliance of participants was excellent, with only two drop-outs due to incomplete proxy responses to the proxy-reported SARC-F.

Characteristics of the patients and proxies from the two cohorts are shown in Table 1 and did not reveal any significant major differences (data not shown). Patients and proxies were almost exclusively Caucasian, with only one Afro-American, one Hispanic, and a limited number of Turkish dyads. Of note, patients from Centres 1 and 2 represented different groups with regard to their functional status, as 93.3% of the patients from Centre 1 compared with only 24.5% of the patients from Centre 2 achieved a SPPB score <9 points and were therefore regarded as having a high risk for sarcopenia-related severe functional impairments. Moreover, the median age at Centre 1 was 81.9 years (range: 63–95 years) in comparison with a median age of 76.6 years (range: 64–90 years) at Centre 2.

In Cohort A, 37.5% of patients had a positive screening result for severe sarcopenia-related functional impairments according to their patient-reported SARC-F score (>3 points), and 40.6% scored <9 points on the SPPB. At T1, 40.6% of all proxies in Cohort A answered the SARC-F questionnaire with the result of a positive screening, and at T2, 50% of proxies evaluated the respective patient as scoring positive (>3 points).

In Cohort B, 30% of patients scored positive for high risk of severe sarcopenia-related functional impairments on the patient-reported SARC-F (>3 points), and 37.5% had a SPPB score <9 points. Fifty-five per cent of all proxies in Cohort B evaluated the respective patient as scoring positive (>3 points) for high risk of severe sarcopenia-related functional impairment at T1 when answering the SARC-F questionnaire at T2.

Performance of the patient-reported and proxy-reported SARC-F for ad-hoc sarcopenia screening

The sensitivity, specificity, PPV, NPV, PLR, and NLR of the patient-reported SARC-F in both cohorts and of the ad-hoc proxy-reported SARC-F in Cohort A are summarized in Table 2. The sensitivity of the different proxy-reported SARC-F modalities was constantly above 0.81, the NPV above 0.86. Overall, sensitivity, specificity, PPV, and NPV were in the range of a suitable screening test.43

Table 2. Descriptive statistic measures of main proxy-reported and patient-reported SARC-F screening Sensitivitya (CI) Specificitya (CI) PPV (CI) NPV (CI) PLR (CI) NLR (CI) Ad-hoc patient-reported SARC-Fcohort A 0.73 (0.52, 0.88) 0.89 (0.74, 0.97) 0.83 (0.61, 0.95) 0.82 (0.66, 0.92) 6.58 (2.54, 17.06) 0.30 (0.16, 0.58) Ad-hoc proxy-reported SARC-Fcohort A 0.81 (0.61, 0.93) 0.89 (0.74, 0.97) 0.84 (0.64, 0.95) 0.86 (0.71, 0.95) 7.27 (2.83, 18.66) 0.22 (0.10, 0.48) Retrospective proxy-reported SARC-F cohort A 0.88 (0.70, 0.98) 0.78 (0.61, 0.90) 0.74 (0.55, 0.88) 0.90 (0.74, 0.98) 3.98 (2.13, 7.45) 0.15 (0.05, 0.44) Ad-hoc patient-reported SARC-Fcohort B 0.53 (0.27, 0.79) 0.84 (0.64, 0.95) 0.67 (0.35, 0.90) 0.75 (0.55, 0.89) 3.33 (1.21, 9.20) 0.56 (0.31, 0.98) Retrospective proxy-reported SARC-Fcohort B 0.87 (0.60, 0.98) 0.64 (0.43, 0.82) 0.59 (0.36, 0.79) 0.89 (0.65, 0.99) 2.41 (1.38, 4.21) 0.21 (0.06, 0.78) CI, 95% confidence interval; NLR, negative likelihood ratio; NPV, negative predictive value; PLR, positive likelihood ratio; PPV, positive predictive value. a SARC-F was regarded as positive with a score >3 points, and the reference to define a high risk for severe sarcopenia-related functional impairments was set at SPPB score <9 points.

In addition, ROC curves for the different SARC-F screening modalities and a SPPB score < 9 points were generated (Figure 2). The AUCs were consistently high, in the range > 0.8. The comparison of the ROC curves revealed no significant differences between the patient-reported and proxy-reported SARC-F versions (Figure 2F).

Receiver operating characteristic (ROC) curves. ROC curves were calculated defining a ‘short physical performance battery’ (SPPB) score <9 points as having a high risk of severe sarcopenia-related functional impairment for (A) ad-hoc patient-reported SARC-F (Cohort A), (B) ad-hoc proxy-reported SARC-F (Cohort A), (C) retrospective proxy-reported SARC-F (Cohort A), (D) ad-hoc patient-reported SARC-F (Cohort B), and (E) retrospective proxy-reported SARC-F (Cohort B). (F) P values for DeLong test comparing areas under the curves (AUC) for the different ROC curves are shown, revealing no significant differences.

Of note, the sensitivity and the PPV were markedly decreased for the ad-hoc patient-reported SARC-F in Cohort B when compared with those in Cohort A (Table 2). Moreover, the AUC for the ad-hoc patient-reported SARC-F in Cohort B was reduced to 0.82 in comparison with 0.92 in Cohort A (Figure 2), although no significant difference was demonstrated (P = 0.195). No obvious reasons for these differences could be identified.

In contrast to the originally proposed SARC-F cut-off of >3 points to define a positive screening result,22 we observed the highest sum of sensitivity and specificity at a higher threshold of >4 points for ROC curves of all patient and proxy subgroups with the exception of the ad-hoc proxy-reported SARC-F in Cohort A (Figure 2). That indicates that the most suitable cut-off for a positive result on the SARC-F for severe impairment would be >4 points for our data.

To assess the interrater reliability between patient and proxy ratings, Cohen's kappa was calculated for agreement between patient-reported and proxy-reported ad-hoc SARC-F > 3 points. Agreement was demonstrated to be substantial [κ = 0.79; CI (0.64, 0.95)].

Correlations of the ad-hoc proxy-reported SARC-F results with the SPPB as well as those of the patient-reported SARC-F with the SPPB in both cohorts were moderate (Table 3). When tested against each other, non-inferiority of the ad-hoc proxy-reported SARC-F/dichotomized SPPB correlation coefficient compared with the patient-reported SARC-F/dichotomized SPPB correlation coefficient could be demonstrated (Table 3). The same was shown for correlations of the ad-hoc proxy-reported and patient-reported SARC-F results with the CRPM and revealed non-inferiority of the ad-hoc proxy-reported SARC-F (Table 3).

Table 3. Summary of non-inferiority analyses CC SARC-F > 3/SPPB < 9 P value 95% CI (r1 − r2) CC SARC-F > 3/chair rise P value 95% CI (r1 − r2) Ad-hoc patient-reported SARC-F (Cohort A) (r1) 0.63 0 [−0.3311, 0.0865] −0.59 0 [−0.0325, 0.3534] Ad-hoc proxy-reported SARC-F (Cohort A) (r2) 0.70 −0.66 Ad-hoc patient-reported SARC-F (Cohort A) (r1) 0.63 0.0065 [−0.3014, 0.2306] −0.59 0.0011 [−0.1385, 0.3894] Retrospective proxy-reported SARC-F (Cohort A) (r2) 0.65 −0.67 Ad-hoc patient-reported SARC-F (Cohort B) (r1) 0.39 0.026 [−0.4774, 0.2313] −0.6 0.268 [−0.5274, 0.2117] Retrospective proxy-reported SARC-F (Cohort B) (r2) 0.49 −0.49 Ad-hoc patient-reported SARC-F (Cohort A) (r1) 0.63 0.0001 [−0.0555, 0.5651]