Diffusion of excellence: evaluating a system to identify, replicate, and spread promising innovative practices across the Veterans health administration

Introduction

The Veterans Health Administration (VHA) Diffusion of Excellence (DoE) program provides a system to identify, replicate, and spread promising practices across the largest integrated healthcare system in the United States. DoE identifies innovations that have been successfully implemented in the VHA through a Shark Tank style competition. VHA facility and regional directors bid resources needed to replicate promising practices. Winning facilities/regions receive external facilitation to aid in replication/implementation over the course of a year. DoE staff then support diffusion of successful practices across the nationwide VHA.

Methods

Organized around the Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE-AIM) Framework, we summarize results of an ongoing long-term mixed-methods implementation evaluation of DoE. Data sources include: Shark Tank application and bid details, tracking practice adoptions through a Diffusion Marketplace, characteristics of VHA facilities, focus groups with Shark Tank bidders, structured observations of DoE events, surveys of DoE program participants, and semi-structured interviews of national VHA program office leaders, VHA healthcare system/facility executives, practice developers, implementation teams and facilitators.

Results

In the first eight Shark Tanks (2016-2022), 3,280 Shark Tank applications were submitted; 88 were designated DoE Promising Practices (i.e., practices receive facilitated replication). DoE has effectively spread practices across the VHA, with 1,440 documented instances of adoption/replication of practices across the VHA. This includes 180 adoptions/replications in facilities located in rural areas. Leadership decisions to adopt innovations are often based on big picture considerations such as constituency support and linkage to organizational goals. DoE Promising Practices that have the greatest national spread have been successfully replicated at new sites during the facilitated replication process, have close partnerships with VHA national program offices, and tend to be less expensive to implement. Two indicators of sustainment indicate that 56 of the 88 Promising Practices are still being diffused across the VHA; 56% of facilities originally replicating the practices have sustained them, even up to 6 years after the first Shark Tank.

Conclusion

DoE has developed a sustainable process for the identification, replication, and spread of promising practices as part of a learning health system committed to providing equitable access to high quality care.

Frontline Clinician Appraisement of Research Engagement: "I feel out of touch with research"

BACKGROUND

Health services research can benefit from frontline clinician input across all stages of research, yet their key perspectives are often not meaningfully engaged.

OBJECTIVE

How can we improve clinician engagement in research?

DESIGN

Convenience sampling and semi-structured interviews followed by descriptive content analysis with an inductive approach, followed by group participatory listening sessions with interviewees to further contextualize findings.

PARTICIPANTS

Twenty-one multidisciplinary clinicians from one healthcare system.

KEY RESULTS

We identified two major themes: perceptions of research (how research fits within job role) and characterizing effective engagement (what works and what does not work in frontline clinician engagement). "Perceptions of Research" encompassed three subthemes: prior research experience; desired degree of engagement; and benefits to clinicians engaging in research. "Characterizing Effective Engagement" had these subthemes: engagement barriers; engagement facilitators; and impact of clinician's racial identity.

CONCLUSIONS

Investing in frontline clinicians as research collaborators is beneficial to clinicians themselves, the health systems that employ them, and those for which they care. Yet, there are multiple barriers to meaningful engagement.

Balancing reality in embedded research and evaluation: Low vs high embeddedness

Embedding research and evaluation into organizations is one way to generate “practice‐based” evidence needed to accelerate implementation of evidence‐based innovations within learning health systems. Organizations and researchers/evaluators vary greatly in how they structure and operationalize these collaborations. One key aspect is the degree of embeddedness: from low embeddedness where researchers/evaluators are located outside organizations (eg, outside evaluation consultants) to high embeddedness where researchers/evaluators are employed by organizations and thus more deeply involved in program evolution and operations. Pros and cons related to the degree of embeddedness (low vs high) must be balanced when developing these relationships. We reflect on this process within the context of an embedded, mixed‐methods evaluation of the Veterans Health Administration (VHA) Diffusion of Excellence (DoE) program. Considerations that must be balanced include: (a) low vs high alignment of goals; (b) low vs high involvement in strategic planning; (c) observing what is happening vs being integrally involved with programmatic activities; (d) reporting findings at the project's end vs providing iterative findings and recommendations that contribute to program evolution; and (e) adhering to predetermined aims vs adapting aims in response to evolving partner needs.

Rapid versus traditional qualitative analysis using the Consolidated Framework for Implementation Research (CFIR)

Background

Qualitative approaches, alone or in mixed methods, are prominent within implementation science. However, traditional qualitative approaches are resource intensive, which has led to the development of rapid qualitative approaches. Published rapid approaches are often inductive in nature and rely on transcripts of interviews. We describe a deductive rapid analysis approach using the Consolidated Framework for Implementation Research (CFIR) that uses notes and audio recordings. This paper compares our rapid versus traditional deductive CFIR approach.

Methods

Semi-structured interviews were conducted for two cohorts of the Veterans Health Administration (VHA) Diffusion of Excellence (DoE). The CFIR guided data collection and analysis. In cohort A, we used our traditional CFIR-based deductive analysis approach (directed content analysis), where two analysts completed independent in-depth manual coding of interview transcripts using qualitative software. In cohort B, we used our new rapid CFIR-based deductive analysis approach (directed content analysis), where the primary analyst wrote detailed notes during interviews and immediately “coded” notes into a MS Excel CFIR construct by facility matrix; a secondary analyst then listened to audio recordings and edited the matrix. We tracked time for our traditional and rapid deductive CFIR approaches using a spreadsheet and captured transcription costs from invoices. We retrospectively compared our approaches in terms of effectiveness and rigor.

Results

Cohorts A and B were similar in terms of the amount of data collected. However, our rapid deductive CFIR approach required 409.5 analyst hours compared to 683 h during the traditional deductive CFIR approach. The rapid deductive approach eliminated $7250 in transcription costs. The facility-level analysis phase provided the greatest savings: 14 h/facility for the traditional analysis versus 3.92 h/facility for the rapid analysis. Data interpretation required the same number of hours for both approaches.

Conclusion

Our rapid deductive CFIR approach was less time intensive and eliminated transcription costs, yet effective in meeting evaluation objectives and establishing rigor. Researchers should consider the following when employing our approach: (1) team expertise in the CFIR and qualitative methods, (2) level of detail needed to meet project aims, (3) mode of data to analyze, and (4) advantages and disadvantages of using the CFIR.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13012-021-01111-5.

The Veterans Health Administration (VHA) Innovators Network: Evaluation design, methods and lessons learned through an embedded research approach

BACKGROUND

Collaboration between researchers, implementers and policymakers improves uptake of health systems research. In 2018, researchers and VHA Innovators Network (iNET) leadership used an embedded research model to conduct an evaluation of iNET. We describe our evaluation design, early results, and lessons learned.

METHODS

This mixed-methods evaluation incorporated primary data collection via electronic survey, descriptive analysis using existing VA datasets (examining associations between facility characteristics and iNET participation), and qualitative interviews to support real-time program implementation and to probe perceived impacts, benefits and challenges of participation.

RESULTS

We developed reporting tools and collected data regarding site participation, providing iNET leadership rapid access to needed information on projects (e.g., target populations reached, milestones achieved, and barriers encountered). Secondary data analyses indicated iNET membership was greater among larger, more complex VA facilities. Of the 37 iNET member sites, over half (n = 22) did not have any of the six major types of VA research centers; thus iNET is supporting VA sites not traditionally served by research innovation pathways. Qualitative findings highlighted enhanced engagement and perceived value of social and informational networks.

CONCLUSIONS

Working alongside our iNET partners, we supported and influenced iNET's development through our embedded evaluation's preliminary findings. We also provided training and guidance aimed at building capacity among iNET participants.

IMPLICATIONS

Embedded research can yield successful collaborative efforts between researchers and partners. An embedded research team can help programs pivot to ensure effective use of limited resources. Such models inform program development and expansion, supporting strategic planning and demonstrating value.

Merging Implementation Practice and Science to Scale Up Promising Practices: The Veterans Health Administration (VHA) Diffusion of Excellence (DoE) Program

INTRODUCTION

The Veterans Health Administration (VHA) Diffusion of Excellence (DoE) program developed and manages a framework for identification, replication, and diffusion of promising practices throughout the nation's largest integrated health care system. DoE identifies promising practices through a "Shark Tank" competition with winning bidders receiving external implementation facilitation. DoE further supports diffusion of successful practices across the VHA.

METHODS

This article presents results of a mixed methods implementation evaluation of DoE, focusing on program reach, program participation and decisions to adopt innovative practices, implementation processes, and practice sustainment. Data sources include practice adoption metrics, focus groups with bidders (two focus groups), observations of DoE events (seven events), surveys of stakeholders (five separate surveys), and semistructured interviews of facility directors, practice developers, implementation teams, and facilitators (133 participants).

RESULTS

In the first four Shark Tank cohorts (2016-2018), 1,676 practices were submitted; 47 were designated Gold Status Practices (practices with facilitated implementation). Motivation for participation varied. Generally, staff led projects targeting problems they felt passionate about, facility directors focused on big-picture quality metrics and getting middle manager support, and frontline staff displayed variable motivation to implement new projects. Approximately half of facilitated implementation efforts were successful; barriers included insufficient infrastructure, staff, and resources. At the facility level, 73.3% of facilities originating or receiving facilitated implementation support have maintained the practice. VHA-wide, 834 decisions to adopt these practices were made.

CONCLUSION

DoE has resulted in the identification of many candidate practices, promoted adoption of promising practices by facility directors, and supported practice implementation and diffusion across the VHA.

Intermediate Diabetes Outcomes in Patients Managed by Physicians, Nurse Practitioners, or Physician Assistants: A Cohort Study

BACKGROUND

Primary care provided by nurse practitioners (NPs) and physician assistants (PAs) has been proposed as a solution to expected workforce shortages.

OBJECTIVE

To examine potential differences in intermediate diabetes outcomes among patients of physician, NP, and PA primary care providers (PCPs).

DESIGN

Cohort study using data from the U.S. Department of Veterans Affairs (VA) electronic health record.

SETTING

568 VA primary care facilities.

PATIENTS

368 481 adult patients with diabetes treated pharmaceutically.

MEASUREMENTS

The relationship between the profession of the PCP (the provider the patient visited most often in 2012) and both continuous and dichotomous control of hemoglobin A1c (HbA1c), systolic blood pressure (SBP), and low-density lipoprotein cholesterol (LDL-C) was examined on the basis of the mean of measurements in 2013. Inverse probability of PCP type was used to balance cohort characteristics. Hierarchical linear mixed models and logistic regression models were used to analyze continuous and dichotomous outcomes, respectively.

RESULTS

The PCPs were physicians (n = 3487), NPs (n = 1445), and PAs (n = 443) for 74.9%, 18.2%, and 6.9% of patients, respectively. The difference in HbA1c values compared with physicians was -0.05% (95% CI, -0.07% to -0.02%) for NPs and 0.01% (CI, -0.02% to 0.04%) for PAs. For SBP, the difference was -0.08 mm Hg (CI, -0.34 to 0.18 mm Hg) for NPs and 0.02 mm Hg (CI, -0.42 to 0.38 mm Hg) for PAs. For LDL-C, the difference was 0.01 mmol/L (CI, 0.00 to 0.03 mmol/L) (0.57 mg/dL [CI, 0.03 to 1.11 mg/dL]) for NPs and 0.03 mmol/L (CI, 0.01 to 0.05 mmol/L) (1.08 mg/dL [CI, 0.25 to 1.91 mg/dL]) for PAs. None of these differences were clinically significant.

LIMITATION

Most VA patients are men who receive treatment in a staff-model health care system.

CONCLUSION

No clinically significant variation was found among the 3 PCP types with regard to diabetes outcomes, suggesting that similar chronic illness outcomes may be achieved by physicians, NPs, and PAs.

PRIMARY FUNDING SOURCE

VA Health Services Research and Development.

ClonEvol: clonal ordering and visualization in cancer sequencing

Background

Reconstruction of clonal evolution is critical for understanding tumor progression and implementing personalized therapies. This is often done by clustering somatic variants based on their cellular prevalence estimated via bulk tumor sequencing of multiple samples. The clusters, consisting of the clonal marker variants, are then ordered based on their estimated cellular prevalence to reconstruct clonal evolution trees, a process referred to as 'clonal ordering'. However, cellular prevalence estimate is confounded by statistical variability and errors in sequencing/data analysis, and therefore inhibits accurate reconstruction of the clonal evolution. This problem is further complicated by intra- and inter-tumor heterogeneity. Furthermore, the field lacks a comprehensive visualization tool to facilitate the interpretation of complex clonal relationships. To address these challenges we developed ClonEvol, a unified software tool for clonal ordering, visualization, and interpretation.

Materials and methods

ClonEvol uses a bootstrap resampling technique to estimate the cellular fraction of the clones and probabilistically models the clonal ordering constraints to account for statistical variability. The bootstrapping allows identification of the sample founding- and sub-clones, thus enabling interpretation of clonal seeding. ClonEvol automates the generation of multiple widely used visualizations for reconstructing and interpreting clonal evolution.

Results

ClonEvol outperformed three of the state of the art tools (LICHeE, Canopy and PhyloWGS) for clonal evolution inference, showing more robust error tolerance and producing more accurate trees in a simulation. Building upon multiple recent publications that utilized ClonEvol to study metastasis and drug resistance in solid cancers, here we show that ClonEvol rediscovered relapsed subclones in two published acute myeloid leukemia patients. Furthermore, we demonstrated that through noninvasive monitoring ClonEvol recapitulated the emerging subclones throughout metastatic progression observed in the tumors of a published breast cancer patient.

Conclusions

ClonEvol has broad applicability for longitudinal monitoring of clonal populations in tumor biopsies, or noninvasively, to guide precision medicine.

Availability

ClonEvol is written in R and is available at https://github.com/ChrisMaherLab/ClonEvol.

Implementing Integrated Mental Health and Chaplain Care in a National Quality Improvement Initiative

This column describes the development, implementation, and outcomes of a quality improvement learning collaborative that aimed to better integrate chaplaincy with mental health care services at 14 participating health care facilities evenly distributed across the U.S. Department of Veterans Affairs and Department of Defense. Teams of health care chaplains and mental health professionals from participating sites sought to improve cross-disciplinary service integration in six key domains: screening, referrals, assessment, communication and documentation, cross-disciplinary training, and role clarification. Chaplains and mental health providers across all facilities at participating sites were significantly more likely post-collaboration to report having a clear understanding of how to collaborate and to report using a routine process for screening patients who could benefit from seeing a professional from the other discipline. Foundational efforts to enhance cross-disciplinary awareness and screening practices between chaplains and mental health professionals appear particularly promising.

Building Capacity in a Rural North Carolina Community to Address Prostate Health Using a Lay Health Advisor Model

Background Prostate cancer is a critical concern for African Americans in North Carolina (NC), and innovative strategies are needed to help rural African American men maximize their prostate health. Engaging the community in research affords opportunities to build capacity for teaching and raising awareness. Approach and Strategies A community steering committee of academicians, community partners, religious leaders, and other stakeholders modified a curriculum on prostate health and screening to include interactive knowledge- and skill-building activities. This curriculum was then used to train 15 African American lay health advisors, dubbed Prostate Cancer Ambassadors, in a rural NC community. Over the 2-day training, Ambassadors achieved statistically significant improvements in knowledge of prostate health and maintained confidence in teaching. The Ambassadors, in turn, used their personal networks to share their knowledge with over 1,000 individuals in their community. Finally, the Ambassadors became researchers, implementing a prostate health survey in local churches. Discussion and Conclusions It is feasible to use community engagement models for raising awareness of prostate health in NC African American communities. Mobilizing community coalitions to develop curricula ensures that the curricula meet the communities' needs, and training lay health advisors to deliver curricula helps secure community buy-in for the information.

Clonal diversity of recurrently mutated genes in myelodysplastic syndromes

Recent studies suggest that most cases of myelodysplastic syndrome (MDS) are clonally heterogeneous, with a founding clone and multiple subclones. It is not known whether specific gene mutations typically occur in founding clones or subclones. We screened a panel of 94 candidate genes in a cohort of 157 patients with MDS or secondary acute myeloid leukemia (sAML). This included 150 cases with samples obtained at MDS diagnosis and 15 cases with samples obtained at sAML transformation (8 were also analyzed at the MDS stage). We performed whole-genome sequencing (WGS) to define the clonal architecture in eight sAML genomes and identified the range of variant allele frequencies (VAFs) for founding clone mutations. At least one mutation or cytogenetic abnormality was detected in 83% of the 150 MDS patients and 17 genes were significantly mutated (false discovery rate ≤0.05). Individual genes and patient samples displayed a wide range of VAFs for recurrently mutated genes, indicating that no single gene is exclusively mutated in the founding clone. The VAFs of recurrently mutated genes did not fully recapitulate the clonal architecture defined by WGS, suggesting that comprehensive sequencing may be required to accurately assess the clonal status of recurrently mutated genes in MDS.

A 3.5 kb deletion in the glycophorin C gene accounts for the Gerbich-negative blood group in Melanesians

The Gerbich-negative blood group types are rare in most populations, but reach appreciable frequencies in certain Melanesian groups in Papua New Guinea. The recent cloning of the human glycophorin C (GPC) gene, that encodes Gerbich (Ge) blood group antigens, has facilitated study of its genetic variants. We have obtained partial genomic clones of a normal GPC gene, for molecular analysis of Ge: -1, -2, -3 types in Melanesians, and have shown that a 3.5 kb deletion in the GPC gene that removes all of exon 3 accounts for at least one Gerbich-negative phenotype in Melanesians. Population distributions of GPC RFLP have shown that the deletion-type GPC is not confined to mainland Papua New Guinea as previously thought, but occurs sporadically in Melanesians from Fiji as well as in Micronesians.

Powerlessness and the pulmonary alveolar edema patient

Everyone wants to maintain control over events in their life. The need for personal control does not end when the patient is hospitalized; instead the patient's need for personal control usually intensifies in critical care situations. The nursing diagnosis of powerlessness is common for most critical care patients, and especially so for the patient experiencing respiratory difficulties such as Pulmonary Alveolar Edema. These authors describe a model of powerlessness which suggests strategies for increasing the patient's control over his or her situation.

Effect of random abnormal cell proportion on specimen classifier performance

A series of papers had analyzed a simplified model of an automated cytology prescreening configuration consisting of a two-class cell classifier followed by a two-class specimen classifier. This has shown, among other things, that the proportion (p) of abnormal cells on an abnormal specimen dictates the number (N) of cells that must be classified before the specimen can be classified with specified accuracy (Anal Quant Cytol, 2:117-122, 1980). It has also shown that if a system designed assuming one fixed value, po, encounters a specimen with a different fixed value, p, then the specimen classifier false negative rate will deviate significantly from the design value, increasing for p < po and vice versa (Cytometry, 2: 155-158, 1981). Using a Gaussian approximation, Timmers and Gelsema (Cytometry, 6:22-25, 1985) extended this to the case where p is a Beta-distributed random variable. They showed that N increases dramatically with the width (coefficient of variation) of the distribution of p. They also concluded that the randomness of p imposes a fundamental lower limit on the specimen false negative rate below which it is impossible to go, even with an error-free cell classifier. In this paper we also extend the basic model to cover the case of random p, but by using an asymptotic expansion (rather than the Gaussian approximation), to develop an expression for N. We show that the limit cited by Timmers and Gelsema is not real, but is actually an artifact of the breakdown of the Gaussian approximation.(ABSTRACT TRUNCATED AT 250 WORDS)

Pulmonary alveolar edema: preventing complications

The critical care nurse uses a wide range of interventions to prevent or reduce complications for the PAE patient. The goal of the interventions are to enhance cardiac performance, improve oxygenation, and decrease myocardial workload.

Nursing management of high permeability pulmonary oedema

In summary, the goal in managing HPPE is to recognise its occurrence and initiate appropriate treatment. While there may be a wide range of possible nursing diagnoses that have application to the HPPE patient, eight essential diagnoses were discussed and outcomes identified: impaired gas exchange; ineffective breathing pattern; ineffective airway clearance; cardiac output, alteration in; fluid volume deficit; infection, potential for; coping, ineffective individual: depression; and powerlessness.

Powerlessness and personal control model applied to the myocardial infarction patient

The MI patient can experience powerlessness through loss of personal control. The feeling of powerlessness can limit the patient's ability to understand the diagnosis of MI, care or decisions necessary to restore health. The MI patient can react by experiencing a sense of physiological, cognitive, environmental and decisional loss of control. Regardless of the specific component of the powerless model, the coronary care nurse diagnoses powerlessness according to defining characteristics. Nursing interventions are organized to facilitate physiological, cognitive, environmental and decisional powerfulness. Research is needed to clarify the MI patient's perception of control or lack of control within each component of the presented model, and to evaluate the effectiveness of nursing interventions created to foster personal control or uncontrol. Research will enable the nurse to scientifically determine strategies and outcomes that correlate with the patient's need for control in the illness situation.

High permeability pulmonary oedema: pathophysiology and mechanisms of injury

Many critically ill patients are at risk for developing HPPE. Since 60% of patients develop HPPE within 24 hours of the pulmonary insult with 11% developing respiratory failure within 72 hours, it is imperative that the critical care nurse understand the pathophysiological responses (Bernard & Bradley, 1986). While the pathophysiological responses are specific, injury to the alveolar-capillary membrane, the mechanisms of injury are diffuse. Knowing the mechanisms can alert health care providers to those patients who are at risk for developing HPPE and more quickly mobilize interventions to alleviate or lessen its occurrence.

HLA-DR and -DQ DNA polymorphisms: new linkage relationships established by RFLP genomic typing in Polynesians and Melanesians

Class II restriction fragment length polymorphisms (RFLPs) of DR beta, DQ beta, and DQ alpha loci were examined in Polynesians of the southwest Pacific and in non-Austronesian-speaking Melanesians from the Papua New Guinean Highlands. Polynesians, previously considered to have a restricted set of HLA-DR antigens, showed class II gene heterogeneity associated with DR2, DR5, DRw6, and DRw8 RFLPs. Furthermore, Melanesians and Polynesians share certain antigens such as DRw6 and DRw8, but the DR beta 2 genes associated with DRw6 and the DQ genes associated with DRw8 are population-specific and show little or no overlap. This study has shown that genetic analysis of closely linked polymorphic genes is a powerful anthropological tool and supports the view that Polynesians represent an independent colonizing group in the Pacific, rather than a group evolved from within Melanesia.

The role of prostacyclin and thromboxane in rat burn and freeze injuries

Deep dermal injuries elicit discrete reaction patterns dependent on the type of injury sustained. Full thickness burn injuries produce an avascular focus of dead and dying tissue surrounded by a peripheral zone of secondary vascular dilation. In contrast, equivalent freeze injuries demonstrate vascular patency both centrally and peripherally. The basis for these differences are unknown. Because of their potent vasoactive and hematologic properties, the presence of two endogenously generated eicosanoids, thromboxane A2 (Tx A2) and prostacyclin (PGI2), were examined in this process. Implanted stainless steel mesh chambers served as an in vivo interstitial collecting reservoir permitting repeated sampling of the wound fluid without tissue disruption. Standard burn and freeze injuries were administered to the skin covering the implanted chambers. The major metabolites of these eicosanoids: 6-keto PGF1 alpha and TxB2 were measured in wound fluid during the first 24 hr following injury. Although both TxB2 and 6-keto PGF1 alpha increased significantly following either injury, treatment with indomethacin did not alter the vascular sequelae despite evident cyclooxygenase inhibition. Latex infusion of whole rats confirmed the considerable difference between these two types of injury, with or without indomethacin. Thus, little evidence was found to support the importance of either TxA2 or PGI2 in the vascular alterations which follow burn or freeze injury.

An expanded cervical cell classification system validated by automated measurements

The commonly used nomenclature for classifying cervical cells was expanded to include more classes, reflecting finer gradations of disease-related morphologic changes. Using a set of subjective visual criteria developed for this purpose, we visually classified 7,000 randomly selected cells and then subjected them to morphologic measurement by digital image analysis. Several of the measurements showed statistically significant differences among all the cell classes, indicating that it is possible to distinguish the finer morphologic gradations incorporated in the new system of cell classes. These same measurements showed a continuous trend of change from class to class along the scale from borderline dysplasia to carcinoma. This is consistent with the notion of a continuous progression of disease development. Finally, we found that those measurements that reflect the visual criteria used in the manual classification were significantly different between classes, indicating that the computer system can successfully quantify many of the important visual criteria.

Endocervical columnar cell atypia coincident with cervical neoplasia characterized by digital image analysis

In the course of classifying uterine cervical epithelial cells for digital image analysis, certain changes in endocervical cells were observed. These changes coincided with the process ongoing in the squamous or metaplasia epithelium. Specifically, in severe dysplasias or carcinomas in situ (CIS), the endocervical nuclei reflected some of the same cytologic changes observed in the dysplastic or CIS squamous cells and yet definitely retained their columnar shape and cytoplasmic quality. This paper deals with almost unexplored area of the endocervical columnar cell in the face of significant cervical neoplasia. Correlation is made between the cytologic criteria observed by the human eye with the light microscope and significant parameters measured by digital image analysis. These measurements suggest that endocervical columnar cells may participate in the dysplastic progression toward CIS.

The effect of abnormal cell proportion on specimen classifier performance

In two previous papers we developed formulas relating the performance (error rates) of a two-class specimen classifier to the performance of a preceding two-class classifier and the number of cells examined (K. R. Castleman and B. S. White, Analytical and Quantitative Cytology 2:117, 1980 and K. R. Castleman and B. S. White, Cytometry 1:156, 1980.). This analysis assumed a certain proportion (p) of abnormal cells on an abnormal specimen. In this paper we examine what happens when a system designed assuming one value of p is presented with a positive specimen having a different abnormal cell proportion. We show that the specimen false negative error rate increases dramatically as p decreases below the design value, and conversely. This suggests that the specimen classification performance of a particular system should be quoted only with reference to the abnormal cell proportion of the specimens used for testing.

The tradeoff of cell classifier error rates

In gynecologic cytodiagnosis it is generally agreed that specimen false negative errors are more serious than false positives. When classifying individual cells, however, it is not obvious how one should adjust the parameters that control the two cell error rates. In the case where a specimen classifier follows a cell classifier, one can calculate the sample size required to achieve specified overall performance. This analysis shows that for the small abnormal cell proportions encountered in cervical cytology, cell false positives should be kept so low that a substantial portion of the abnormal cells are missed.

Optimizing cervical cell classifiers

In an automated prescreening system where a cell classifier and a specimen classifier operate in cascade, the false-positive and false-negative error rates of each classifier can be traded off to obtain the best overall performance. It is usually desirable to keep the specimen false-negative rate below the false-positive rate. An analysis of the classifier cascade shows that, in contrast, the cell classifier should have its false-positive rate much lower than its false-negative rate. A procedure is presented for selecting the best operating point on the ROC curve of the cell classifier. This minimizes the sample size required to achieve prescribed specimen error rates.