Part XVI · Disparities, access, and health systems · Chapter 92

Implementation, policy, and quality of care

Efficacy is established in trials. Whether anyone receives it is a separate empirical question with its own methods.

1 · Translating evidence into practice

A trial establishes that an intervention can work under specified conditions. Whether it does work in routine care is a different question, and it has its own literature, its own methods and its own failure modes.

Implementation frameworks exist because the reasons an effective intervention fails to spread are numerous and overlapping. The Consolidated Framework for Implementation Research organises them into five domains: characteristics of the intervention itself, the outer setting, the inner setting, the characteristics of the individuals involved, and the process of implementation¹. Its value is diagnostic. Before choosing an implementation strategy, a team has to say which domain the obstruction sits in, because strategies aimed at one domain do nothing to another.

The distinction that matters most in practice is between effectiveness and implementation outcomes. Effectiveness outcomes are clinical: survival, recurrence, toxicity. Implementation outcomes are different quantities: adoption, reach, fidelity, sustainment. An intervention can have excellent effectiveness and zero reach, and a programme reporting only the first will describe a success that no patient experienced.

Two examples in this part make the point concretely. Clinical breast examination downstaged breast cancer in a randomised trial and the effect on mortality did not reach significance². Patient navigation improved timely care overall and produced no benefit at all in the first 90 days³. Neither is a negative result. Both are results about where in a pathway an intervention acts, which is an implementation question rather than an efficacy one.

2 · Multidisciplinary care models and referral systems

The multidisciplinary team is the standard organisational form for breast cancer care, and what it actually delivers is worth separating into three distinct functions.

It pools expertise, so a decision reflects surgical, oncological, radiological and pathological input. It standardises, so two patients with the same disease at the same institution receive the same recommendation. It creates an audit point, because a documented team decision can be measured against a guideline.

Those three functions are not equally well evidenced, and only the third is straightforwardly measurable. A team meeting that reviews sixty cases in an hour has performed the third function and possibly not the first.

The referral system around the team is what determines who reaches it. Its performance is measurable at three transitions: primary care to diagnostic service, diagnostic service to team review, and team decision to treatment start. Each transition has its own loss rate.

The loss rates are not small. Among 46 185 abnormal screening mammograms recommended for biopsy, 34.6% were not resolved within 30 days and 12.2% were not resolved within 90 days⁴. The identity of the screening facility accounted for part of the racial difference in that delay, which means the transition is a property of the institution rather than only of the patient⁴.

3 · Guideline adherence and justified variation

Not all variation from a guideline is error, and a quality programme that treats adherence as the outcome will suppress the variation that should exist.

Three categories of departure need separating, and only one of them is a quality problem.

Justified variation follows from a patient characteristic the guideline did not address. Comorbidity, frailty, competing risk of death and informed patient preference all produce departures that are correct. A guideline derived from trials that excluded patients over 75 does not govern an 85-year-old with cardiac failure.

Unjustified variation reflects clinician habit, local convention or resource availability presented as clinical judgement. This is the target of a quality programme.

Evidentiary variation arises where the guideline itself rests on weak evidence, and two reasonable clinicians reach different conclusions from the same data. Enforcing adherence here converts uncertainty into false consensus.

The distinction cannot be made from an adherence rate alone, which is why a bare percentage is uninformative. It requires recorded reasons for departure. A service that documents the reason can separate the three categories. A service that records only concordance cannot, and will improve its measured adherence by discouraging justified departures.

There is also a floor below which the question does not arise. In a population-based cohort of 2998 patients with invasive breast cancer, 62% of those eligible had no 21-gene assay performed⁵. That is not a variation in judgement about what the assay shows. It is the absence of the input on which the judgement would be made.

4 · Quality indicators and outcome measurement

An indicator is a measurement convention. It is chosen for reproducibility across institutions rather than because it marks the point where outcome changes, and reading it as the second is the same error described for assay thresholds in HER2 heterogeneity.

Indicators divide into three types and they behave differently under pressure.

Structure indicators describe what a service has: a linear accelerator, an accredited laboratory, a specialist nurse. They are easy to verify and weakly connected to outcome.

Process indicators describe what a service does: the proportion receiving an assay, the proportion starting treatment within a target interval. They are actionable, and they are the type most vulnerable to being met without being achieved.

Outcome indicators describe what happens to patients: survival, recurrence, patient-reported outcome. They are the point of the exercise and they require risk adjustment, which imports every problem of the model used to adjust.

The Global Breast Cancer Initiative indicators are a clean worked example of process measurement designed deliberately, since all three are health system quantities rather than clinical ones⁶. Their weakness is instructive too. An attempt to measure them across 21 Asian countries found that over 60% had no published national data on staging or on the intervals from presentation to diagnosis and from diagnosis to treatment. Only five of the 21 reported data on treatment completion. Definitions of delayed diagnosis and of treatment completion varied between countries⁷.

5 · Policy as natural experiment, coverage expansion and access reform

Coverage reform is one of the few interventions in this part that can be studied with something approaching a control group. Where a policy is adopted by some jurisdictions and not others at a known date, a difference-in-differences design compares the change in adopting jurisdictions with the change in non-adopting ones.

The design is not randomised and its assumption is specific. It requires that, absent the policy, the two groups' outcomes would have moved in parallel. Jurisdictions choose whether to adopt, and they choose for reasons that may also affect outcomes.

The clearest breast cancer result concerns Medicaid expansion under the Affordable Care Act. Among 9322 patients diagnosed with de novo stage IV breast cancer in states that expanded Medicaid in January 2014, survival was compared before and after. Before expansion, White patients had better overall survival than patients from racial and ethnic minority groups, with an adjusted hazard ratio of 1.22 and limits of 1.10 to 1.35. After expansion that difference was no longer observed, with an adjusted hazard ratio of 0.96 and limits of 0.86 to 1.08⁸.

That is a survival disparity narrowing after a change in coverage and not in treatment. It is the strongest available evidence that a substantial part of the gap is payment rather than biology, which is the argument of Decomposing an outcome gap into biology, access, and delivery arriving through a policy door.

The result is also narrower than it is usually reported, and the qualification matters. A separate analysis of breast, colon, lung and prostate cancer found that Medicaid expansion produced no change in receipt of stage-appropriate definitive treatment for racial and ethnic minority patients compared with non-expansion states. It did produce an increase of 3.62 percentage points in the proportion treated within 30 days, with limits of 1.63 to 5.61, and no change in the proportion treated within 90 days. At minority-serving hospitals there was no change in any of the three measures⁹.

Coverage changed timing at the patient level. It did not change what treatment was received, and it did not change anything at the facilities serving the most patients from minority groups. Both findings are about the same policy.

6 · Health system accountability and public reporting

Public reporting assumes a causal chain, and each link in it can fail independently.

The chain runs: a measure is published, someone reads it, that reader changes behaviour, and the change improves outcome. The intended readers are patients choosing a provider, referrers choosing where to send, and institutions responding to reputational pressure.

Patient choice is the weakest link for breast cancer specifically. Choice requires an alternative within reach, and the geography set out in Access, affordability, and financial toxicity means many patients have one accessible centre. A published measure changes nothing for a patient with no second option.

Institutional response is the strongest link and also the most vulnerable to gaming, through the denominator and threshold mechanisms described in Quality indicators and outcome measurement.

Risk adjustment is where accountability schemes most often break, and the failure has a predictable direction. A centre treating more socially disadvantaged patients will have worse crude outcomes. If the adjustment model includes social factors, the centre is held to a lower standard and disparity is normalised into the measurement. If it excludes them, the centre is penalised for its case mix and the incentive is to avoid those patients. Both choices are defensible and they cannot both be made.

The honest resolution is to report both. A crude figure describes what happened to the population. An adjusted figure describes the institution's contribution. A scheme reporting only one of the two is answering only one of the two questions.

7 · Equity-focused implementation research

An implementation framework asks why an effective intervention did not spread. An equity-focused one asks why it spread unevenly, which is a different question requiring different variables.

The Health Equity Implementation Framework integrates determinants from implementation science with determinants from health disparities research, and adds the clinical encounter itself as a level of analysis¹⁰. Its distinguishing feature is that it separates barriers that are general from barriers that are specific to the disadvantaged group. A preliminary application to hepatitis C treatment found both kinds in the same population. Poor care coordination affected everyone. Accounts from peers of racial discrimination in the same health system affected one group¹⁰.

The methodological consequence is the one to carry into practice. An intervention evaluated on its average effect can widen a gap while improving the mean. It will do so whenever uptake is easier for the already advantaged, which is the default for any intervention requiring time, travel, literacy or connectivity.

This gives a concrete design requirement rather than a principle. Report the effect by subgroup as a prespecified outcome, not as a secondary analysis. Patient navigation is the model. Its benefit was greatest at the centres with the longest delays under usual care, which was reported as part of the main result rather than discovered afterwards³. That single finding tells a service where to deploy the intervention, and a whole-population mean would not have.

8 · Interventions that did not work, and why that matters

Null and partial results in implementation are informative in a way that null efficacy results are not, because they identify which link in a chain was already intact.

Four from this part are worth holding together.

Patient navigation produced no benefit during the first 90 days of care, where most patients are resolved. The benefit appeared between 91 and 365 days³. The intervention did not improve routine care. It recovered patients who were being lost, which is a smaller and more specific claim than the one usually made for it.

Medicaid expansion increased the proportion of patients from racial and ethnic minority groups treated within 30 days by 3.62 percentage points. It changed nothing about receipt of definitive treatment, nothing at 90 days, and nothing at minority-serving hospitals⁹. Removing a payment barrier does not build capacity at the facilities where capacity is short.

Clinical breast examination screening downstaged disease substantially and did not produce a significant mortality reduction in the prespecified analysis². Downstaging converts to mortality benefit only where the treatment that follows is available and completed, which is the sequential logic of the Global Breast Cancer Initiative pillars in Global breast cancer care.

Protocol-assigned treatment in a randomised trial did not eliminate the outcome gap. In TAILORx there was no substantial difference in 21-gene recurrence score by race, and treatment was assigned by protocol. Black women still had a hazard ratio of 1.41 for overall survival against White women in the whole population, with limits of 1.05 to 1.90¹¹. Standardising what is prescribed does not standardise what is experienced.

References

All 11 references

1. Damschroder LJ, Aron DC, Keith RE, Kirsh SR, Alexander JA, Lowery JC. Fostering implementation of health services research findings into practice: a consolidated framework for advancing implementation science. Implement Sci 2009 4:50. PMID 19664226
2. Mittra I, Mishra GA, Dikshit RP, et al. Effect of screening by clinical breast examination on breast cancer incidence and mortality after 20 years: prospective, cluster randomised controlled trial in Mumbai. BMJ 2021 372:n256. PMID 33627312
3. Freund KM, Battaglia TA, Calhoun E, et al. Impact of patient navigation on timely cancer care: the Patient Navigation Research Program. J Natl Cancer Inst 2014 106:dju115. PMID 24938303
4. Lawson MB, Bissell MCS, Miglioretti DL, et al. Multilevel factors associated with time to biopsy after abnormal screening mammography results by race and ethnicity. JAMA Oncol 2022 8:1115-1126. PMID 35737381
5. Dunn MR, Li D, Emerson MA, et al. A latent class assessment of healthcare access factors and disparities in breast cancer care timeliness. PLoS Med 2024 21:e1004500. PMID 39621782
6. Boucheron P, Zietsman A, Anele A, et al. Measuring the WHO Global Breast Cancer Initiative Pillars' key performance indicators in Sub-Saharan Africa: experience in the African Breast Cancer-Disparities in Outcomes hospital-based cohort study. EClinicalMedicine 2025 81:103104. PMID 40034567
7. Ong SK, Haruyama R, Yip CH, et al. Feasibility of monitoring Global Breast Cancer Initiative Framework key performance indicators in 21 Asian National Cancer Centers Alliance member countries. EClinicalMedicine 2023 67:102365. PMID 38125964
8. Malinowski C, Lei X, Zhao H, Giordano SH, Chavez-MacGregor M. Association of Medicaid expansion with mortality disparity by race and ethnicity among patients with de novo stage IV breast cancer. JAMA Oncol 2022 8:863-870. PMID 35389432
9. Nguyen DD, Paciotti M, Marchese M, et al. Effect of Medicaid expansion on receipt of definitive treatment and time to treatment initiation by racial and ethnic minorities and at minority-serving hospitals: a patient-level and facility-level analysis of breast, colon, lung, and prostate cancer. JCO Oncol Pract 2021 17:e654-e665. PMID 33974827
10. Woodward EN, Matthieu MM, Uchendu US, Rogal S, Kirchner JE. The health equity implementation framework: proposal and preliminary study of hepatitis C virus treatment. Implement Sci 2019 14:26. PMID 30866982
11. Albain KS, Gray RJ, Makower DF, et al. Race, ethnicity, and clinical outcomes in hormone receptor-positive, HER2-negative, node-negative breast cancer in the randomized TAILORx trial. J Natl Cancer Inst 2021 113:390-399. PMID 32986828