Professional Documents
Culture Documents
VINCENT MOR, PHD,' LINDA LALIBERTE, MS,t JOHN N. MORRIS, PHD,S AND MO-IAEL WIEMANN, MD§
The Karnofsky Performance Status Scale (KPS)is widely used to quantify the functional status of cancer
patients. However, limited data exist documenting its reliability and validity. The K P S is used in the
National Hospice Study (NHS) as both a study eligibility criterion and an outcome measure. As part
of intensive training, interviewers were instructed in and tested on guidelines for determining the KPS
levels of patients. After 4 months of field experience, interviewers were again tested based on narrative
patient descriptions. The interrater reliability of 47 NHS interviewers was found to be 0.97. The construct
validity of the K P S was analyzed, and the K P S was found to be strongly related (P < 0.001) to two
other independent measures of patient functioning. Finally, the relationship of the K P S to longevity (r
= 0.30) in a population of terminal cancer patients documents its predictive validity. These findings
suggest the utility of the KPS as a valuable research tool when employed by trained observers.
Cancer 53:2002-2007, 1984.
2002
No. 9 THEKARNOFSKY PERFORMANCE
STATUSSCALE - Mor et a/. 2003
Able to cany on normal activity; no special care needed. 100 Normal, no complaints, no evidence of disease.
90 Able to carry on normal activity, minor signs or symptoms of
disease.
80 Normal activity with effort. some signs or symptoms of
disease.
Unable to work, able to live at home and care for most 70 Cares for self, unable to carry on normal activity or to do
personal needs, varying amount of assistance needed. work.
60 Requires occasional assistance from others but able to care for
most needs.
50 Requires considerable assistance from others and frequent
medical care.
Unable 10 care for self, requires institutional or hospital care 40 Disabled, requires special care and assistance.
or equivalent, disease may be rapidly progressing.
30 Severely disabled, hospitalization indicated, death not
imminent.
20 Very sick, hospitalization necessary, active supportive
treatment necessary.
10 Moribound.
0 Dead.
admitted to the NHS between August 1, 1981 and July prepared by NHS staff. Formal training began with an
31, 1982 who had died by November 30, 1982 and who orientation to the basic principles of cancer biology,
had a KPS score of 50 or less at the initial interview. symptoms, epidemiology, and therapy. Led by a clinical
oncologist and an oncologic nurse, this 3-hour orientation
Methods included practical exercises in which the interviewers
demonstrated their ability to abstract selected items of
The NHS interviewer makes initial contact with the information from the medical records of terminal cancer
NHS participant 3 to 10 days following admission to the patients.
hospice. At this initial contact, the interviewer spends Following this orientation, the interviewers took part
approximately 1 hour with the patient and the patient’s in a 2-hour training session conducted by the same clinical
family, gathering information on pain, symptoms, sat- team. The KPS index used by the NHS interviewers is
isfaction with care, satisfaction with life, and services being shown in Table I . Operational specifications were de-
utilized. During this period, the interviewer has the o p veloped to provide objective guidelines for nonclinical
portunity to observe the patient’s subtle dependencies staff in the use of this judgement rating scale. The goal
and interactions within the existing support networks to of the training was to enable interviewers to rate and
amve at a KPS assessment based on that single visit. compare patient profiles at subtly different performance
Additional selected functional assessment information is levels, generalizing this experience to include all types of
gathered either independently by hospice clinical staff at patients.
the time of the patient’s admission o r from the patient’s Following their introduction to the overall rating pro-
family at the initial contact. The interviewer continues cess,interviewers were tested with a seriesof brief narrative
to visit the patient and the patient’s family regularly until descriptions of patients and their respective KPS scores
the patient’s death. until there was thorough agreement on the basis for each
particular rating.
Training
Results
The NHS interviewers were selected primarily on the
Reliability
basis of their prior interviewing and research experience.
Consideration was also given to their familiarity with As noted above, although the Kps is widely used by
medical and insurance/billing terminology. There were clinicians, recent evidence suggests that ratings frequently
9 nurses among the 47 interviewers; the remaining 8 1% are not comparable.” In research applications, however,
had very limited clinical experience in a medical setting. raters using the KPS have been found to achieve at least
Prior to the training period, as a general background a moderate level of interrater agreement.I I Reliability re-
the interviewers were asked to become familiar with ter- fers to the extent to which objects (patients) can be cat-
minology contained in a medical glossary which had been egorized, scored, or ordered in the same way by different
2004 CANCERMay I 1984 Vol. 53
TABLE2. Performance Level of Activities ical status is far more complicated. By comparing the
of Daily Living by Karnofsky Status KPS with other discrete observable variables which con-
Karnofsky performance status ceptually or theoretically tap the same measurement do-
(%) main, its construct validity can be e~a1uated.l~
Functional In the NHS, a series of independent physical func-
variables 10 20 30 40 50 Total
tioning measures is available with which to test the con-
Continence struct validity of the Karnofsky Performance Index. At
Unable 28.6 28.3 15.3 7.1 .9 12.0
Wlhelp 50.0 55.4 59.4 50.6 27.0 50.6 the time of intake, hospice staff complete a series of de-
Independent 21.4 16.3 25.3 42.3 72.2 31.4 tailed functional assessments on the subjects based on
Transfering inlout the general model of the Katz ADL Index.14The patient’s
of bed/chair
Unable 42.9 45.7 20.3 7.1 0.0 16.0 primary care person completes an overall physical QL
Wlhelp 57.1 44.6 60.2 58.4 27.0 52.5 (quality of life) assessment at the initial and each sub-
Independent 0.0 9.8 19.5 34.5 73.0 31.5 sequent interview contact. While the three sets of as-
Walking
Unable 50.0 57.6 36.4 18.4 .9 27.4 sessments are not identical in content or intent, they do
Wlhelp 50.0 33.7 48.3 49.8 32.2 44.6 measure facets of the same underlying phenomenon, pa-
independent 0.0 8.7 15.3 31.8 67.0 28.0 tient performance. Consequently, there should be a strong
Dressing
Unable 71.4 64.1 52.9 21.0 1.7 35.4 relationship between these other two sets of functional
Wlhelp 28.6 31.5 40.2 60.7 49.6 41.7 measures and the KPS. The descriptors of the various
Independent 0.0 4.3 6.9 18.4 48.7 17.0 KPS levels suggest not only a physical performance di-
Bathing
Unable 57.1 60.9 52.5 24.0 4.3 36.0 mension but also medical care needs. Symptom presence
Wfhelp 42.9 35.9 41.0 59.6 56.6 49.4 and extent of disease, therefore, should conceptually cor-
independent 0.0 3.3 6.5 16.5 39.1 14.6 relate with the KPS. Data regarding the presence and
Climbing stairs
Unable 78.6 87.0 71.3 51.7 10.4 57.0 severity of symptoms, including pain, are obtained from
Wlhelp 21.4 10.9 24.1 40.8 61.7 34.2 the patient during the initial interview. The patient’s
Independent 0.0 2.2 4.6 1.5 27.8 8.8 medical record was used to record the location and num-
Mobility outside home
Unable 78.6 83.7 63.6 43.1 16.5 51.8 ber of metastases.
W/help 21.4 15.2 34.5 52.4 66.1 43.1 The performance level for activities of daily living for
Independent 0.0 1.1 1.9 4.5 17.4 5.1 patients at different KPS scores is shown in Table 2. The
Overall 1.9 12.3 34.8 35.6 15.4 100.0 chi-square tests for all categories indicate that differences
in performance level for each KPS score are significant
No. = 685.
W: with. (Pc 0.001), with the proportion of patients able to func-
tion independently increasing as the patients’ KPS scores
increase. As the KPS scores decrease, the proportion of
assessors. To test whether consensus among interviewers patients unable to perform these activities with or without
was achieved, a written quiz similar to that given during assistance increases drastically, indicating a severely dis-
the training process was administered to interviewers after abled population at a KPS of <30.
4 months of field work. Forty-seven interviewers were A severity index was constructed from the functional
asked to rate 17 narrative examples of patients at widely variables listed in Table 2. For each variable, a score of
varying KPS levels. The results of this quiz were analyzed 0 was assigned if the patient could perform the activity
to determine interrater reliability. Interviewers’ ratings independently, a score of 1 was assigned if assistance was
were compared using Cronbach’s coefficient alpha, which necessary, and a 2 was assigned if the patient was unable
treats judges as items in a scale and the narrative scenarios to perform the activity. The scores range from 0 (most
as observations. This approach was supplemented by a functional) to 14 (least functional) and have been distri-
variant of the intraclass correlation suggested by Winer. ’* butionally categorized. Table 3 presents the severity index
Both approaches yielded interrater reliability coefficients for patients at each KPS level. The chi-square test indicates
of over 0.97 of a maximum of 1.00. This value is statis- -=
significant differences among categories (P 0.00 1) and
tically significant beyond the 0.00 1 probability level. a very strong relationship between increasing KPS scores
and higher functional status (contingency coefficient
Construct Validity
= 0.49). Over 50% of the cases in the two worst KPS
The KPS is a complex, synthetic measure of a patient’s levels are in the worst severity category, while 40% of the
activity levels. It is a multidimensional construct reduced patients with a KPS of 50 are in the “most functional”
to a unidimensional scale. One of the reasons for the severity category.
popularity of the KPS is precisely its concise approach The relationship between the KPS and the physical
to classifying and ranking patients whose individual clin- QL is given in Table 4. There are significant differences
No. 9 THEKARNOFSKY STATUSSCALE
PERFORMANCE - Mor ef al. 2005
TABLE5. Analysis of Variance Relating Patients’ Karnofsky with KPS scores of 50 live longer than 36 days. Kendall’s
Performance Status Score to Longevity tau associated with this table is 0.30, which is significant
~ _ _ _ _ _ _ ~
Karnofsky Mean Standard Median beyond the 0.001 level. Interestingly, the proportion of
score longevity deviation longevity No. patients at each KPS level found in the mid-range lon-
gevity group is fairly comparable.
10 17.6 20.5 9.5 13
20 27.1 29.4 17.8 84 These analyses substantiate the predictive validity of
30 45.7 43.4 31.9 239 the KPS when applied to a sample of terminal cancer
40 64. I 52.4 46.7 244 patients. While the relationship is highly significant, its
50 72.0 52.8 59.7 105
predictive accuracy and therefore case-specific clinical
Overall 53.5 48.9 36.6 685 applicability is limited. Clearly, however, the lower KPS
F-ratio: 19.2. levels are much more likely to be “death-imminent.”
P < 0.001.
“Eta squared”: 0.09.
Longevity: days alive; KPS: Karnofsky performance status. Discussion
The wide use of the KPS among physicians and par-
To complement the ANOVA results, the ordinal KPS ticularly oncologists suggests it is a meaningful measure
score was regressed on survival in days. The purpose of of a patient’s functional status. However, for this or any
this analysis was to further examine the predictive qualities similar measure to be meaningful, it is essential that those
of the KPS. Confirming the ANOVA approach, a zero using it be systematically trained to assure a common
order correlation of 0.30 was obtained, signifying an R2 perspective. The training module designed for the current
of 0.09 which is significant at beyond the 0.001 level. evaluation could be readily adapted by other researchers
Each increase in one KPS level, e.g., 20 to 30, yields an and by clinicians working in a practice or research setting.
increase of approximately 15 days of survival. Regression Our finding of high levels of interrater reliability among
analyses also reveal that a higher proportion of prediction nonclinicians following exposure to training attests to the
errors (standardized residuals greater than or -2.0) + feasibility of using the KPS in systematic research efforts.
occur among patients with KPS levels of 40 and above; The functional variables adapted from Katz ADL In-
over 6%of cases with KPS scores of 40 or 50 were “errors,” dex14 and used to test the construct validity of the KPS
whereas only 4% of KPS level 30 and 2% of KPS level are objective and quantifiable. Thus, the strong relation-
20 were errors. Furthermore, significant errors of pre- ship between these variables and the KPS as shown in
diction were uniformly in the direction of longer lives. Table 2 suggests that although the KPS is a subjective
This is largely a function of the skewed longevity distri- rating, it is based on objective factors. The strong rela-
bution for this sample seen in the median being lower tionship obtained between the KPS and a single item
than the mean. physical QL judgement made by patients’ primary care
In view of these findings, a categoric form of longevity persons indicates the consistency with which the construct
was constructed. The proportion of “short-lived” and of performance status can be measured. This construct
“long-lived” patients at each KPS level was examined. validity, therefore, supports the view that the KPS is a
Three longevity groups were created: ( 1) 18 days or less; reliable measure of functional status.
(2) 29 to 36 days; and (3) more than 36 days. The results Perhaps most important is the finding that the KPS is
are presented in Table 6. As can be seen, the majority a predictor of length of survival in terminally ill cancer
of patients with KPS scores of 10 or 20 die within 18 patients. The predictive relationship that emerged, while
days of hospice admission, while most (70.4%) of those accounting for only 9% of the variation in longevity, is
consistent with other researchers’ efforts to predict lon-
gevity in chronically ill and aged population^.^^-^' The
TABLE6. Karnofsky Performance Status By Category more limited relationship to survival reported by Yates
of Patient Longevity
and associates” was presumably due to greater hetero-
KPS Level (7%) geneity in his sample. By limiting the sample to patients
with confirmed metastatic disease with an acknowledgedly
Patient longevity 10 20 30 40 50 Total
poor outcome, the relationship between performance sta-
1-18 d 71.4 52.2 29.1 13.9 8.7 24.2 tus and longevity becomes apparent. However, even in
19-36 d 21.4 25.0 27.6 26.6 20.9 25.8 this population, the predictive accuracy attained is in-
37 d or more 7.1 22.6 43.3 59.6 70.4 50.0
sufficient to guide clinical practice and decision-making.
Total 1.9 12.3 34.8 35.6 15.4 100.0 Perhaps the most effective use of the KPS when reliably
No. = 685. applied is as a stratifyingmeasure in research, particularly
KPS: Karnofsky Performance Status. in clinical trials in which patient survival constitutes an
No. 9 THEKARNOFSKY
PERFORMANCE
STATUSSCALE M o r et al. 2007
outcome of interest or is related to other outcomes of for the study of chemotherapy of cancer in man: Comparative therapeutic
interest. In view of the strong relationship between KPS trial of nitrogen mustard and triethylene thiophosphoramide. J C'hron
D ~ s1960; I1:7-33.
and survival, even randomized clinical trials could gain 6. Eastern Cooperative Oncology Group. Functional assessment scale.
considerable power by using the KPS as a study eligibility Rubin P, ed. Clinical Oncology: A Functional Approach. New York:
criterion and as a stratifying variable during analyses. On American Cancer Society, 1983; 9 1.
7. Yates JW, McKegney FP, Kun LE. A comparative study of home
the basis of the high short-term mortality among patients nursing care of patients with advanced cancer. In: American Cancer
in the lowest KPS groups, it might be reasonable to ex- Society. Proceedings of the American Cancer Society's National Con-
clude them from studies requiring some minimal exposure ference on Human Values and Cancer 1980; 207-2 18.
8. Cohen MN, Makuch R, Johnston-Early A el al. Laboratory pa-
time. The widespread use, ease of measurement, and gen- rameters as an alternative to performance status in prognostic stratifi-
eral acceptance of the measure make the KPS a reasonable cation of patients with small cell cancer. Cunccv Treat Rep 1981;
candidate as a stratifying variable as long as the reliability 65(3-4): 187-195.
9. Gutman RA, Stead WW, Robinson RR. Physical activity and
of the raters is assured. employment status of patients on maintenance dialysis. N Engl J Med
Future investigations into the predictors of survival 1981; 304(6):309-313.
time among terminal cancer patients will explore other 10. Huthinson TA, Boyd NF, Feinstein AR, Gorda A. tiollomhy D,
Rowat B. Scientific problems in clinical scales, as demonstrated in the
factors, i.e.,concurrent disabling diseases, prior treatment Karnofsky Index of Performance Status. J Chmn Dis 1979; 32:661-
patterns, etc., that, together with the KPS, can refine such 666.
predictions. I I . Yates JW, Chalmer B, McKegney FP. Evaluation ofpatients with
advanced cancer using the Karnofsky Performance Status. Cancer 1980;
REFERENCES 45(8):222&2224.
12. Winer BJ. Statistical Principles in Experimental Design. New
I . Hansen HH, Dombernowsky P, Hirsch FR. Staging procedures York: McGraw-Hill, 197 I; 283-296, 5 18-525.
and prognostic features in small cell anaplastic bronchogenic carcinoma. 13. Nunnally JC. Psychometric Theory. New York: McGraw-Hill,
Semin Oncol 1978; 5:280-287. 1967.
2. Maurer LH. Pajak TF. Prognostic factors in small cell carcinoma 14. Katz S, Ford AB, Moskowitz RW, Jackson BA. Jare MW. Studies
of the lung: A cancer and leukemia group B study. Cancer Trear Rep of illness in the aged. The Index of ADL: A standardized measure of
I98 I ; 65:767-774. biological function. J Am Med Assoc 1963; 185(12):914-919.
3. Karnofsky DA, Abelmann WH, Craver LF, Burchenal JH. The 15. Melzack R. The McGill Pain Questionnaire: Major properties
use of nitrogen mustards in the palliative treatment of cancer. Cancer and scoring methods. fain 1975; 1(3):277-299.
1948; 1(4):634-656. 16. Linn BS, Linn MW. Gurel L. Measurement of physical impair-
4. Karnofsky DA, Burchenal JH. The clinical evaluation of che- ment and its relationship to aging and death. In: Palmore E, Jemers FC.
motherapeutic agents in cancer. In: MacLeod CM, ed. Evaluation of eds. Prediction of Life Span. Lexington, Massachusetts: DC Heath CO.,
Chemotherapeutic Agents. New York: Columbia University Press, 1949, 1971.
I9 1-205. 17. Pfeiffer E. Survival in old age: Physical, psychological, and social
5 . Zubrod CG, Schneiderman M, Frei E el al. Appraisal of methods correlates of longevity. J Am Geriafr Soc 1970; 18(4):273-285.