Bottom-Up Costing

Bottom-Up Costing from Patient-Level Resource Use

Bottom-up costing builds the cost of a treatment from the patient upwards. The analyst records what each patient in a sample actually consumed, such as staff minutes, theatre time, medicines and bed-days, multiplies each quantity by a unit cost and adds the results. Top-down costing runs the other way, starting from an organisation's recorded expenditure and sharing it out among services or patients. The choice matters in economic evaluation because the two routes can give different costs for the same service, and because bottom-up data show how costs vary between patients. Even a bottom-up study usually needs top-down information to value bundled resources, and the NICE manual (PMG36) has its own position on which costs to use.

Two separate choices about where data start and how detailed they are

Špacírová and colleagues, building on a taxonomy by Tan and colleagues, describe two questions that every costing method answers. The first is the degree of disaggregation, from micro-costing of detailed items to gross costing of aggregated ones. The second is the method of valuation. Top-down methods apportion expenditure collected at cost-centre level down to units of activity, whereas bottom-up methods first identify the resources used by individual patients and then value them with unit costs.

Crossing the two gives four categories.

The framework explains a frequent confusion. Published studies do not always keep bottom-up costing and micro-costing apart, but a study can follow patients and record only bed-days and visits (bottom-up gross costing), while activity-based costing is classed by Špacírová and colleagues as a top-down micro-costing method because it allocates indirect expenditure through activities and cost drivers. They add that bottom-up costing needs no cost drivers, since resources are attributed directly to each patient followed, and suggest reserving the term for studies of patients. Xu and colleagues report that micro-costing studies variously describe themselves as bottom-up, ingredient-based or give no name at all.

The three steps of identifying, measuring and valuing

Špacírová and colleagues describe bottom-up methods as following three phases, usually in order. The analyst selects a sample of patients and identifies the resources to record along their care pathway for a set period; measures each resource in natural units such as staff time, units of medicine or minutes of room use; and assigns a monetary value, usually by multiplying the units used by a unit cost. For one patient the cost is as follows.

$$C_i = \sum_{j=1}^{J} q_{ij} \, p_j$$

where $C_i$ is the cost for patient $i$, $q_{ij}$ is the quantity of resource $j$ used by patient $i$ in natural units, $p_j$ is the unit cost of resource $j$ and $J$ is the number of resource items recorded. The sample mean $\bar{C} = \frac{1}{n}\sum_{i=1}^{n} C_i$, where $n$ is the number of patients, is the figure usually carried into a model.

The perspective decides which items belong in the inventory and which unit costs apply. For economic evaluation the unit value should in principle reflect the social opportunity cost of the resource, although many analyses take a narrower perspective for practical reasons. In the NICE reference case, costs should relate to resources under the control of the NHS and personal social services, valued at prices relevant to them, with evidence that resource use and cost data were identified systematically (PMG36, section 4.4.1).

Collecting the quantities

Frick's review of micro-costing studies lists ten ways in which quantity data were gathered, namely administrative databases at single facilities, insurer administrative data, forms applied across several sites, an expert panel, surveys or interviews of providers, patient chart review, direct observation, personal digital assistants, programme operation logs and diaries. Frick stresses that the need for precision has to be weighed against its cost. An expert panel using a modified Delphi approach collects no data on specific patients but, like any recall method, risks recall bias. Direct observation, as in a colonoscopy study where a nurse accompanied every study patient and recorded times, medicines and equipment, records resource use as care is delivered. The price of detail shows in a US study of residents in nursing units using hand-held computers, which needed a week of staff preparation for 48 hours of nurse data.

In a chlamydia screening study, data were collected on high and low workload days and at the start, middle and end of the study, because early operation may not reflect efficient delivery. Frick notes that micro-costing studies are often run at one or a few centres because of the effort involved, which makes transfer to other settings harder, and recommends reporting quantities and prices separately so that readers can apply local values. Xu and colleagues found that 86.2% of the micro-costing studies they appraised did not report unit costs and resource quantities separately.

Unit costs and the bundled-resource problem

For primary resources such as staff time or a dose of a medicine, the price the organisation pays is the obvious starting value, although Špacírová and colleagues point out that salaries and medicine prices are often set by national agreement rather than by markets. The harder case is the composite resource. A day on a hospital ward or a radiograph bundles direct items with shared costs such as catering, laundry, ward supervision, equipment and buildings that cannot be traced to one patient.

To value such bundles the analyst has to return to top-down methods or use published tariffs or prices. When a top-down cost per day is used, it should exclude anything already recorded patient by patient. If nurse contact minutes were measured directly, nursing should not be counted again in the day rate. Plugging in values from national unit cost databases saves time, can represent providers across a country better than a single-site figure when compiled from a representative sample, and makes studies more comparable. Tariffs and public prices may not reflect the economic cost of the resources involved. Overheads in general need a cost allocation rule, because no method can classify every cost as direct.

Accuracy, precision and the cost of detail

Špacírová and colleagues separate accuracy (whether the estimate suits the question and is well founded) from precision (whether repeated measurement would give consistent results, which improves with sample size). They note that many authors consider bottom-up methods more accurate than top-down methods and some call bottom-up micro-costing the gold standard, but argue that no costing method deserves that label, because bottom-up costing can measure individual resource use accurately yet cannot value composite resources without top-down inputs, and its precision depends on sample size and is reduced by dropout or missing data.

The empirical comparisons support a selective approach. Tan and colleagues costed four hospital services in Dutch general hospitals and concluded that restricting bottom-up micro-costing to components with a large effect on total cost, namely labour and inpatient stay, would probably give reliable estimates. Xu and colleagues summarise three studies in which bottom-up costing gave more accurate estimates for labour-intensive services. Chapko and colleagues compared two US Veterans Affairs costing systems for 14,915 patients. Correlations ranged from 0.24 for outpatient encounters to 0.77 for inpatient admissions and 0.85 for total annual cost, and agreement increased with aggregation. Špacírová and colleagues argue that both systems were in fact top-down, differing only in how overheads were allocated.

What NICE PMG36 says about micro-costing

The NICE technology appraisal manual describes healthcare resource groups (HRGs) as standard groupings of clinically similar treatments, whose national average unit costs come from the annual mandatory collection of reference costs from NHS organisations in England. Using them can reduce the need for local micro-costing, which the manual defines as costing each individual component of care related to a technology (section 4.4.9).

HRG data may not be appropriate when the new technology and its comparator fall in the same HRG, or when the mean cost does not reflect the resource use of the new technology; micro-costing studies may then be more appropriate, and all relevant costs such as testing, follow-up, monitoring, staffing, facilities and training should be included (section 4.4.10). Costs from earlier years should be inflated to present values with an index suited to the perspective, such as the NHS cost inflation index (section 4.4.12).

Worked example with three surgical admissions (illustrative)

All numbers are illustrative. A hospital costs a planned surgical procedure from the NHS provider perspective by following three patients. Patient C has a wound complication that lengthens the stay. In GBP, surgeon time costs 120 per hour, theatre time excluding the surgeon 20 per minute and the implant kit 400 per kit. A ward day excluding nurse contact time costs 300 and recorded nurse contact time 50 per hour.

1. Record and value each patient's resources.

For patient B the formula gives

$$C_B = 1.5 \times 120 + 90 \times 20 + 1 \times 400 + 4 \times 300 + 6 \times 50 = 3880$$

where each term is a quantity multiplied by its unit cost in GBP, in the order of the table rows.

2. Summarise. Total cost is 2,470 + 3,880 + 3,970 = 10,320, so the mean is 10,320 / 3 = 3,440 per admission. The range from 2,470 to 3,970 is information that a single top-down cost per admission cannot show.

3. See where the cost sits. Theatre time (1,200 + 1,800 + 1,200 = 4,200) and ward days (600 + 1,200 + 1,800 = 3,600) make up 7,800 of 10,320, or 75.6%. Directly recorded staff time comes to 420 + 900 = 1,320 (12.8%) and kits to 1,200 (11.6%). Measurement effort is best spent on theatre minutes and length of stay, in line with Tan and colleagues' advice to concentrate on labour and inpatient stay.

4. Check for double counting. Suppose the analyst instead takes a top-down ward-day cost of 375 that already includes average ward nursing (300 + 1.5 hours × 50), and still adds the recorded nurse contact time. Nursing is then counted twice at 75 per day over the 12 ward days, an extra 12 × 75 = 900, or 300 per patient. The mean rises to 3,740, overstating the correct 3,440 by 8.7%.

5. Interpret. Patient C cost 3,970 / 2,470 = 1.61 times as much as patient A, and 1,200 of the 1,500 difference comes from four extra ward days. A technology that shortened stays would show its saving clearly in these data. Whether a ward day valued at its average cost of 300 is actually released as cash depends on capacity utilisation and on the marginal cost of a bed-day, not on the average.

Where bottom-up costing fits and where it misleads

Frick names three situations where micro-costing should be the first choice, namely a new intervention with no average cost yet, a study of variation within a procedure, and analyses that include non-market goods such as volunteer or carer time. He also notes that individual-level data are relevant for describing the precision of results in a cost-effectiveness analysis. Špacírová and colleagues suggest a bottom-up approach for capturing local variation and a top-down approach for generalising over a whole service population.

Several errors recur. Treating a bottom-up figure as exact ignores the top-down inputs inside every bundled unit cost. Small samples followed for a short period can miss rare, costly events such as complications. Costs measured at a single site may not transfer, for example where staff qualifications differ. Frick reports one colonoscopy study that adjusted staff costs for operation below capacity, while others reviewed either assumed full capacity or did not say. Studies labelled bottom-up that in fact allocate ledger costs through drivers belong in the top-down group, and reporting only totals without separate quantities and prices prevents readers from re-costing the results. Applying the same patient-level logic to all the costs of a disease in a population is the subject of bottom-up cost-of-illness studies, a separate entry.

Sources

• Chapko MK, Liu CF, Perkins M, Li YF, Fortney JC, Maciejewski ML. Equivalence of two healthcare costing methods: bottom-up and top-down. Health Economics. 2009;18(10):1188-1201. https://doi.org/10.1002/hec.1422
• Frick KD. Microcosting quantity data collection methods. Medical Care. 2009;47(7 Suppl 1):S76-S81. https://doi.org/10.1097/MLR.0b013e31819bc064
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. Section 4.4, Evidence on resource use and costs. https://www.nice.org.uk/process/pmg36/chapter/economic-evaluation-2 (accessed 5 October 2026).
• Špacírová Z, Epstein D, García-Mochón L, Rovira J, Olry de Labry Lima A, Espín J. A general framework for classifying costing methods for economic evaluation of health care. European Journal of Health Economics. 2020;21(4):529-542. https://doi.org/10.1007/s10198-019-01157-9
• Tan SS, Rutten FF, van Ineveld BM, Redekop WK, Hakkaart-van Roijen L. Comparing methodologies for the cost estimation of hospital services. European Journal of Health Economics. 2009;10(1):39-45. https://doi.org/10.1007/s10198-008-0101-x
• Xu X, Lazar CM, Ruger JP. Micro-costing in health and medicine: a critical appraisal. Health Economics Review. 2021;11:1. https://doi.org/10.1186/s13561-020-00298-5