Barthel Index

Barthel Index in Stroke Rehabilitation and Economic Evaluation

The Barthel Index records how much help a person needs with basic self-care and getting about, and in health economics it is best known as the functional outcome of stroke, rehabilitation and intermediate care trials. Its total carries no preference weights, so a Barthel score cannot be multiplied by time to give QALYs. The index reaches a cost-utility analysis only by mapping to the EQ-5D, by defining dependency states in a model, or by serving as a natural unit in a cost-effectiveness ratio. Each route rests on choices that change results, and Sulter and colleagues found that acute stroke trials had used cut-off scores anywhere between 50 and 95 to define a favourable outcome, so "independent" in one study is not the same state as "independent" in another.

Ten items and two scoring scales

Mahoney and Barthel published the index in 1965 in the Maryland State Medical Journal. It has ten items, which Sulter and colleagues divide into a self-care group (feeding, grooming, bathing, dressing, bowel and bladder care and toilet use) and a mobility group (ambulation, transfers and stair climbing). Each item is rated by the level of help needed, and the items carry different weights.

On the original scale the item scores move in steps of 5. Sulter and colleagues describe a maximum of 100, indicating full physical independence, and a lowest score of 0, representing a totally dependent bedridden state. Because every step is a multiple of 5, dividing a 0 to 100 total by 5 gives the 20-point form used by Collin, Wade and colleagues at the Rivermead Rehabilitation Centre in Oxford. Mapi Research Trust, which distributes the instrument, lists a ten-item version scored 0 to 20 alongside the original.

Several modified versions exist. Shah, Vanclay and Cooper expanded the number of categories used to record improvement in each function and proposed new scoring guidelines, testing them on 258 first-time stroke patients admitted for inpatient rehabilitation in Brisbane in 1984. They reported that the modified scoring was more sensitive and more reliable than the original, with internal consistency of 0.90 against 0.87. Wade and Collin, by contrast, argued that the standard Barthel should be adopted as a common measure of physical disability and advised against modified versions. In a 2011 review Quinn, Langhorne and Stott proposed the ten-item scale scored 0 to 100 in 5-point increments as the uniform stroke trial version.

The Maryland State Medical Society holds the copyright. Mapi Research Trust states that permission is needed to modify the Barthel or to use the index for commercial purposes, which matters for industry-sponsored trials that plan to adapt the wording.

Raters, proxies and reliability

The Barthel Index records what a person does, and the patient, a nurse, a relative or a trained assessor can complete the ratings. When a clinician or trained assessor rates the patient, the index works as a clinician-reported outcome, and when a relative supplies the ratings it is an observer report.

Collin and colleagues compared four ways of obtaining scores in 25 patients, namely self-report, a trained nurse who had worked with the patient for at least one shift, and separate testing by two skilled observers. Total scores agreed closely across the four methods, a difference of 4 points out of 20 was likely to reflect a genuine difference, and asking an informed nurse or relative was as reliable as direct testing and quicker. A meta-analysis by Duffy and colleagues pooled 10 studies with 543 participants and found excellent interobserver reliability for standard administration, with a weighted kappa of 0.93 (95% confidence interval 0.90 to 0.96).

Proxy completion is one reason the index interests economists working on stroke. Van Exel, Scholte op Reimer and Koopmanschap noted that many patients in the acute phase are physically or mentally unable to report their own quality of life, which leaves gaps in EQ-5D data that can bias cost-effectiveness results, whereas the Barthel can be completed by a proxy.

How Barthel scores reach QALYs and cost-effectiveness ratios

The Barthel total is an ordinal count of independence, and no value set turns it into a utility. It is therefore a non-preference-based measure, and an analyst who wants a QALY for a cost-utility analysis has three options, each with its own assumptions.

Mapping Barthel scores to the EQ-5D

A mapping function predicts EQ-5D utility from Barthel responses. Van Exel, Scholte op Reimer and Koopmanschap examined the relationship between the two measures in 598 patients from three stroke-service experiments in the Netherlands, with 6-month follow-up, and reported a plausible and significant relationship. Their aim was to fill in quality of life for patients who could not complete the EQ-5D themselves.

Kaambwa, Billingham and Bryan used 1,189 patients in UK intermediate care to compare ordinary least squares, censored least absolute deviations (CLAD) and multinomial logistic models. Models that treated EQ-5D as a continuous outcome with Barthel dimensions as predictors performed better, CLAD had the lowest mean absolute error and ordinary least squares the lowest root mean squared error, and the multinomial logistic model did worst on both. They concluded that EQ-5D can reasonably be predicted from the Barthel and recommended the CLAD model.

In NICE's guideline on stroke rehabilitation in adults (NG236, 2023), the economic model for botulinum toxin A to reduce spasticity notes that EQ-5D data were reported in only two randomised trials of the treatment. It lists the Barthel Index, the Modified Ashworth Scale, the Disability Assessment Scale and a numeric rating scale for pain among the outcomes considered to enable health economic modelling. The final model estimated QALYs from Modified Ashworth Scale responder data, with EQ-5D values for responders and non-responders, so the Barthel was considered but not used.

Dependency states defined by Barthel cut-offs

A model can instead define health states by Barthel bands and attach a utility and a cost to each. The band edges are a modelling choice. Sulter and colleagues, reviewing acute stroke efficacy trials, found favourable outcome defined by Barthel sum scores anywhere between 50 and 95. They suggested that poor outcome might be easier to define and proposed death, institutionalisation due to stroke, a modified Rankin score above 3 or a Barthel score below 60 as poor-outcome end points.

The Barthel as a natural unit

An evaluation can instead report cost per point of Barthel improvement, or cost per patient reaching independence, in a cost-effectiveness analysis. Such ratios cannot be compared with a cost per QALY threshold, and a cost per point depends on whether the 20-point or the 100-point scale was used.

Illustrative worked example of mapping a Barthel gain to a cost per QALY

All numbers in this example are illustrative, including the mapping coefficients, which are not taken from any published function. A rehabilitation programme gives a mean Barthel score at 6 months of 13 on the 20-point scale, against 11 under standard care, at an incremental cost of GBP 1,600 per patient. The analyst uses a linear mapping.

$$u = \alpha + \beta B$$

where $u$ is the predicted EQ-5D utility, $B$ is the Barthel total on the 0 to 20 scale, $\alpha$ is the intercept, set at −0.10, and $\beta$ is the utility gained per Barthel point, set at 0.04.

1. Predicted utilities. Standard care gives 0.04 × 11 = 0.44, so the utility is −0.10 + 0.44 = 0.34. The programme gives 0.04 × 13 = 0.52, so the utility is −0.10 + 0.52 = 0.42.

2. QALY gain. The utility gain is 0.42 − 0.34 = 0.08. If the gain lasts one year, the QALY gain is 0.08 × 1 = 0.08.

3. Cost per QALY. The ratio is 1,600 ÷ 0.08 = 20,000, or GBP 20,000 per QALY.

4. Cost per Barthel point. The gain is 2 points on the 20-point scale, so the cost is 1,600 ÷ 2 = 800, or GBP 800 per point. The same gain is 10 points on the 100-point scale, giving 1,600 ÷ 10 = 160, or GBP 160 per point.

The cost per QALY can be set against a cost-effectiveness threshold, while the cost per point has no threshold and changes fivefold with the choice of scale. The QALY figure rests entirely on the mapping slope and on how long the gain is assumed to last. With a slope of 0.03 the utility gain would be 0.03 × 2 = 0.06 and the ratio 1,600 ÷ 0.06, about GBP 26,667 per QALY, which is why mapping uncertainty belongs in the sensitivity analysis.

Floor and ceiling effects and other limits

The Barthel was built to describe dependence in hospital and rehabilitation patients, and its limits follow from that purpose. Quinn and colleagues concluded that the index is a valid measure of activities of daily living with acceptable reliability, but that its sensitivity to change is limited at the extremes of disability through floor and ceiling effects.

• Ceiling. A person can score the maximum while still unable to shop, cook or work, so a ceiling effect hides further gains among milder patients, and a mapped utility for full Barthel independence will not distinguish them.
• Narrow content. The ten items cover physical self-care and mobility. Cognition, mood, pain and social participation lie outside the index, so a treatment whose benefits fall there will look ineffective on the Barthel and on any utility mapped from it.
• Unequal steps. The scores are ordinal, so a 10-point change near the bottom of the scale need not equal a 10-point change near the top, and mean differences should be read with that in mind.
• Scale confusion. A Barthel change of 2 is small on the 0 to 100 total and large on the 20-point form, and the expanded categories of the Shah version do not convert exactly to either, so effect sizes taken from trials need their scale checked before pooling or mapping.
• Arbitrary cut-offs. Dichotomising at 50 or at 95 produces different proportions of "good outcome" from the same data, and the choice changes the number of patients occupying each state in a model.

Barthel Index, Katz Index and modified Rankin Scale

Several measures of daily function compete for the same place in stroke and ageing studies, and the choice affects what a model can represent. The Barthel and the Katz index both measure basic activities of daily living, but they differ in structure.

The Katz ADL Index rates six functions as dependent or independent and summarises them as a count or a letter grade, while the Barthel weights ten items, adds stair climbing and walking, and grades partial help. The modified Rankin Scale is a single global grade of disability after stroke, and Sulter and colleagues used a modified Rankin score above 3 beside a Barthel score below 60 in their proposed definition of poor outcome. Because the modified Rankin Scale grades the whole person and the Barthel grades specific tasks, a stroke model can use the first to define broad dependency states and the second to describe function within them, provided the two are not counted as separate benefits.

Sources

• Collin C, Wade DT, Davies S, Horne V. The Barthel ADL Index: a reliability study. International Disability Studies. 1988;10(2):61-63. doi:10.3109/09638288809164103
• Duffy L, Gajree S, Langhorne P, Stott DJ, Quinn TJ. Reliability (inter-rater agreement) of the Barthel Index for assessment of stroke survivors: systematic review and meta-analysis. Stroke. 2013;44(2):462-468. doi:10.1161/STROKEAHA.112.678615
• Kaambwa B, Billingham L, Bryan S. Mapping utility scores from the Barthel index. European Journal of Health Economics. 2013;14(2):231-241. doi:10.1007/s10198-011-0364-5
• Mahoney FI, Barthel DW. Functional evaluation: the Barthel Index. Maryland State Medical Journal. 1965;14:61-65.
• Mapi Research Trust. Barthel Index (BI). ePROVIDE instrument record, consulted 2026. https://eprovide.mapi-trust.org/instruments/barthel-index
• National Institute for Health and Care Excellence. Stroke rehabilitation in adults (NG236). Economic model: cost-utility analysis of botulinum toxin A to reduce spasticity. Published 18 October 2023. https://www.nice.org.uk/guidance/ng236/evidence/economic-model-costutility-analysis-of-botulinum-toxin-a-to-reduce-spasticity-pdf-13192131712
• Quinn TJ, Langhorne P, Stott DJ. Barthel index for stroke trials: development, properties, and application. Stroke. 2011;42(4):1146-1151. doi:10.1161/STROKEAHA.110.598540
• Shah S, Vanclay F, Cooper B. Improving the sensitivity of the Barthel Index for stroke rehabilitation. Journal of Clinical Epidemiology. 1989;42(8):703-709. doi:10.1016/0895-4356(89)90065-6
• Sulter G, Steen C, De Keyser J. Use of the Barthel index and modified Rankin scale in acute stroke trials. Stroke. 1999;30(8):1538-1541. doi:10.1161/01.STR.30.8.1538
• van Exel NJA, Scholte op Reimer WJM, Koopmanschap MA. Assessment of post-stroke quality of life in cost-effectiveness studies: the usefulness of the Barthel Index and the EuroQoL-5D. Quality of Life Research. 2004;13(2):427-433. doi:10.1023/B:QURE.0000018496.02968.50
• Wade DT, Collin C. The Barthel ADL Index: a standard measure of physical disability? International Disability Studies. 1988;10(2):64-67. doi:10.3109/09638288809164105