Karnofsky Performance Status

Karnofsky Performance Status in Oncology Trials and Treatment Rules

The Karnofsky scale puts a number on how far cancer limits a patient's everyday life, and it still decides who receives some treatments. A clinician places the patient at one of eleven levels in steps of ten, with 100 for normal health and 0 for death. The NICE guideline on brain tumours and brain metastases (NG99) uses a score of 70 or more as a condition for several treatment recommendations, so the score sets the size of the treated population and with it the budget and the QALYs that follow. In NICE guidance the scale appears most clearly in glioma, where it sits beside the ECOG grades used in most cancer models and cannot be swapped for them without error.

Origins as a measure of chemotherapy benefit

Timmermann's history of the scale credits it to David Karnofsky and Joseph Burchenal, and Péus and colleagues cite a 1949 chapter by the two authors, in a book on the evaluation of chemotherapeutic agents, as its origin. Timmermann describes the scale as a standardisation device that helped quantify effects of chemotherapeutic agents that were less easily measured than survival time.

That purpose matters for its economic use. Timmermann reports that the scale was used only sporadically in the 1950s and 1960s, mostly around experimental chemotherapy, and that interest in measuring quality of life emerged only around 1970. Performance scales such as Karnofsky's were then reinterpreted, retrospectively, as simple quality of life indices. The scale was never designed to capture health-related quality of life as a patient experiences it, and it carries no preference weights.

Eleven levels and how a score is reached

The treating clinician assigns the level that best matches the patient's everyday functioning and need for care. The result is a clinician-reported outcome. Timmermann describes eleven stages, with 100 standing for normal health and 0 for death, and each lower level records a greater limitation. Level 10 is the last level above death, and level 0 records death itself.

The levels are ordered categories and not measurements on an interval scale, an instance of ordinal scaling. A fall of 10 points near the top of the scale and a fall of 10 points near the bottom describe very different changes, so averaging scores across patients or treating the difference between two means as a clinical effect needs caution.

Structured questions to make ratings consistent

Ratings depend on how the clinician gathers information, and researchers have tried to standardise that step. Schag, Heinrich and Ganz asked 293 cancer patients to complete a questionnaire on physical and psychosocial difficulties while physicians rated them on the scale, and used a subsample of 75 patients to assess agreement between raters. Using regression they identified seven behaviour-based questions that predicted the score, covering weight loss, weight gain, reduced energy, difficulty walking, driving, grooming and working part time, and proposed an interview based on them.

Péus and colleagues later proposed a short algorithm that reaches a score from two or three questions. Both aim to make ratings more consistent between clinicians by narrowing the room for individual judgement.

Karnofsky thresholds in NICE guidance and trial eligibility

In economic work the scale matters most as a rule that defines who is treated. NICE's guideline NG99, published in July 2018 and last updated in January 2021, uses a Karnofsky score of 70 or more as a condition in its recommendations for glioma. For newly diagnosed grade 4 glioma (glioblastoma), radiotherapy with concomitant and adjuvant temozolomide is offered to people aged around 70 or under with a score of 70 or more, and best supportive care alone is to be considered for people aged around 70 or over with a score under 70. For recurrent high-grade glioma the guideline lists the Karnofsky score among the factors to take into account.

A threshold of this kind fixes the eligible population, and the eligible population drives a budget impact analysis. It also acts as a subgroup boundary. If a trial enrolled only patients at or above a cut point, a subgroup analysis cannot say much about patients below it, and an economic model extended to them rests on extrapolation.

Combining Karnofsky and ECOG evidence

Most modern cancer trials and models record the ECOG Performance Status instead, so analysts sometimes need to move between scales. Verger, Salamero and Conill compared both scales in 150 consecutive cancer patients, derived a regression-based relation in half of the sample and tested it in the other half. They published transformation tables with 95% and 66% confidence intervals and found the scales highly correlated, but predictions for an individual patient carried substantial error. Translation was hardest at lower performance status, where scores spread widely.

A model that relabels a Karnofsky-defined population in ECOG terms, or applies utilities estimated by ECOG grade to Karnofsky groups, therefore inherits that error. The conversion is an assumption to state and test in a sensitivity analysis, not a fixed equivalence.

Karnofsky groups inside a cost-effectiveness model

A model can use the score in three places, and each needs different evidence. As a baseline characteristic, the share of the cohort above and below a cut point should match the population the decision covers, which for an NHS decision means local or registry data and not only the trial's enrolment. As a covariate, the score can adjust survival or progression equations fitted to trial data, which requires patient-level data that record it. As a basis for utilities, a Karnofsky group has to be linked to scores from a preference-based instrument collected in the same patients, because the Karnofsky level itself cannot be read as a QALY weight.

The choice of grouping is itself a modelling decision. Groups that follow the threshold used in guidance, such as 70 or more against under 70, keep the model aligned with the treatment rule, while finer groups use more of the scale but need more data in each group to estimate costs, utilities and survival.

Worked example of rater variation at a treatment threshold (illustrative)

All numbers in this example are illustrative. A cancer network sees 200 people a year with newly diagnosed glioblastoma who meet the age condition, and treatment is offered only to those with a Karnofsky score of 70 or more. Treatment costs GBP 9,500 more per patient than best supportive care and is assumed to gain 0.30 QALYs per treated patient.

$$B = N_e \times c$$

where $B$ is the annual incremental budget, $N_e$ is the number of eligible patients and $c$ is the incremental cost per treated patient.

1. Ratings by the first clinician. The first clinician scores 140 patients at 70 or above, so the budget is 140 × 9,500 = 1,330,000, or GBP 1,330,000 a year.

2. Ratings by a second clinician. A second clinician scores 12 of those 140 patients at 60, and scores 4 patients rated 60 by the first clinician at 70. The eligible number becomes 140 − 12 + 4 = 132, and the budget is 132 × 9,500 = 1,254,000, or GBP 1,254,000.

3. Difference. The budgets differ by 1,330,000 − 1,254,000 = 76,000, which equals the 8 fewer eligible patients multiplied by the cost, 8 × 9,500 = 76,000. The QALYs gained differ by 8 × 0.30 = 2.4 QALYs.

The cost per QALY is the same under both raters, because each treated patient is assumed to cost and gain the same, so a one-step disagreement near the threshold changes the scale of spending and benefit, not the ICER. That holds only if patients just above and just below 70 benefit equally. If patients near the cut point gain less than those scored higher, the rater who admits more of them lowers the average benefit, and the ratio rises. A budget impact estimate for NHS practice that takes the eligible share from a single set of ratings therefore understates its uncertainty, and a sensitivity range on the eligible number reflects the misclassification that rater variation can cause.

How the Karnofsky scale differs from nearby measures

The Karnofsky scale shares its purpose with other clinician-rated functional scales. It differs from them in how finely it divides function and in the diseases where it is used, and the table compares it with the measures most often used beside it.

The Karnofsky and ECOG scales describe the same ground at different grain, and the modified Rankin Scale plays a similar role for stroke. Neither Karnofsky nor ECOG scale yields a health state utility value, so a cost-utility model that groups patients by Karnofsky score must attach utilities from a preference-based instrument to each group.

Limitations and common misreadings

The score's simplicity hides several weaknesses that matter when it is used to set eligibility or to define model groups. Most of them come from the fact that one clinician's judgement stands in for a measured quantity.

• Agreement between raters. Conill, Verger and Salamero reported a correlation of 0.76 between physicians rating the Karnofsky score of the same 100 cancer patients. A correlation of that size leaves room for one-step disagreements at a threshold.
• Patients and clinicians disagree. Schnadig and colleagues found that patients and physicians disagreed on the Karnofsky score for 67.1% of 1,636 patients with advanced lung or colorectal cancer, against 56.6% for the ECOG grade, and in adjusted analysis disagreement on the Karnofsky score was associated with worse survival, with a hazard ratio of 1.16.
• Reading it as quality of life. Timmermann shows that the scale was devised to standardise judgements of chemotherapy benefit and was reinterpreted as a quality of life index only later, so it is a weak substitute for a patient-reported measure.
• Treating a percentage as a proportion of health. A score of 50 does not mean half of full health and cannot be used as a QALY weight.
• Assuming a fixed ECOG equivalent. Verger and colleagues found substantial individual error when converting scores, so a single conversion table is a modelling assumption and a source of measurement error.

Sources

• Conill C, Verger E, Salamero M. Performance status assessment in cancer patients. Cancer. 1990;65(8):1864-1866. doi:10.1002/1097-0142(19900415)65:8<1864::AID-CNCR2820650832>3.0.CO;2-U
• Karnofsky DA, Burchenal JH. The clinical evaluation of chemotherapeutic agents in cancer. In MacLeod CM, ed. Evaluation of Chemotherapeutic Agents. New York, Columbia University Press; 1949. pp. 191-205.
• National Institute for Health and Care Excellence. Brain tumours (primary) and brain metastases in over 16s (NG99). Published 11 July 2018, last updated 29 January 2021. https://www.nice.org.uk/guidance/ng99/chapter/recommendations
• Péus D, Newcomb N, Hofer S. Appraisal of the Karnofsky Performance Status and proposal of a simple algorithmic system for its evaluation. BMC Medical Informatics and Decision Making. 2013;13:72. doi:10.1186/1472-6947-13-72
• Schag CC, Heinrich RL, Ganz PA. Karnofsky performance status revisited: reliability, validity, and guidelines. Journal of Clinical Oncology. 1984;2(3):187-193. doi:10.1200/JCO.1984.2.3.187
• Schnadig ID, Fromme EK, Loprinzi CL, et al. Patient-physician disagreement regarding performance status is associated with worse survivorship in patients with advanced cancer. Cancer. 2008;113(8):2205-2214. doi:10.1002/cncr.23856
• Timmermann C. 'Just give me the best quality of life questionnaire': the Karnofsky scale and the history of quality of life measurements in cancer trials. Chronic Illness. 2013;9(3):179-190. doi:10.1177/1742395312466903
• Verger E, Salamero M, Conill C. Can Karnofsky performance status be transformed to the Eastern Cooperative Oncology Group scoring scale and vice versa? European Journal of Cancer. 1992;28A(8-9):1328-1330. doi:10.1016/0959-8049(92)90510-9