Strengthening the Center for Medicare and Medicaid Innovation's Approach to Constructing Alternative Payment Models.

The Center for Medicare and Medicaid Innovation (CMMI) seeks to develop evidence-based alternative payment models (APM) to improve health care quality and reduce costs, but its performance in achieving these goals has been mixed. In October 2021, CMMI released its Innovation Strategy Refresh to highlight challenges faced by payment models and suggest new strategic approaches for the upcoming decade. While a welcome recast of organizational goals, the Refresh leaves space for how CMMI will address persistent issues. These include how CMMI can best engage physicians and patients in APMs, minimize conflicting incentives among APMs, reduce selection bias in model participation, and, ultimately, transition away from the fee-for-service framework that underlies much of Medicare reimbursement. This article provides guidance to CMMI's vision by examining challenges within CMMI's strategy for model building and offering solutions to mitigate these issues. These strategies include engaging beneficiaries in APM incentives, expanding operational flexibility to improve clinical behaviors (e.g., waivers), rectifying issues with conflicting model incentives, building voluntary short-term and mandatory long-term incentives to mitigate selection bias, and transitioning to an overriding population-based model to constrain net costs. Policy Points The Center for Medicare and Medicaid Innovation (CMMI) seeks to develop evidence-based alternative payment models (APM) to improve care quality and reduce health care cost, but its performance in achieving these goals has been mixed. In October 2021, CMMI released a "strategic refresh" of its goals but left space for how persistent issues to model development would be addressed. We propose strategies to engage physicians and patients in APMs, minimize conflicting incentives among APMs, reduce selection bias in model participation, and, ultimately, transition away from the fee-for-service framework that underlies much of Medicare reimbursement.

Imaging versus Intervention in Managing Small Unruptured Intracranial Aneurysms: A Cost-Effectiveness Analysis.

OBJECTIVE: Current guidelines recommend active surveillance with serial magnetic resonance angiography (MRA) for management of small, asymptomatic unruptured anterior circulation aneurysms (UIAs). We sought to determine the cost-effectiveness of active surveillance compared to immediate surgery. METHODS: We developed a Markov cost-effectiveness model simulating patients with small (<7 mm) UIAs managed by active surveillance via MRA, immediate surgery, or watchful waiting. Inputs for the model were abstracted from the literature and used to construct a comprehensive model following persons from diagnosis to death. Outcomes were quality-adjusted life-years (QALYs), lifetime medical costs (2015 USD), and incremental cost-effectiveness ratios (ICERs). Cost-effectiveness, deterministic, and probabilistic sensitivity analyses were performed. RESULTS: Immediate surgical treatment was the most cost-effective management strategy for small UIAs with ICER of USD 45,772 relative to active surveillance. Sensitivity analysis demonstrated immediate surgery was the preferred strategy, if rupture rate was >0.1%/year and if the diagnosis age was <70 years, while active surveillance was preferred if surgical complication risk was >11%. Probabilistic sensitivity analysis demonstrated that at a willingness-to-pay of USD 100,000/QALY, immediate surgical treatment was the most cost-effective strategy in 64% of iterations. CONCLUSION: Immediate surgical treatment is a cost-effective strategy for initial management of small UIAs in patients <70 years of age. While more costly than MRA, surgical treatment increased QALY. The cost-effectiveness of immediate surgery is highly sensitive to diagnosis age, rupture rate, and surgical complication risk. Though there are a wide range of rupture rates and complications associated with treatment, this analysis supports the treatment of small, unruptured anterior circulation intracranial aneurysms in patients <70 years of age.

Impact of Marijuana Legalization on Opioid Utilization in Patients Diagnosed with Pain.

BACKGROUND: Given efforts to reduce opioid use, and because marijuana potentially offers a lower-risk alternative for treating chronic pain, there is interest in understanding the public health impact of marijuana legalization on opioid-related outcomes. OBJECTIVE: Assess the impact of recreational and medical marijuana legalization on opioid utilization among patients receiving pharmacotherapy for pain. DESIGN: Retrospective claims-based study of commercially insured patients continuously eligible for pharmacy and medical benefits from July 8, 2014 to June 30, 2017. Index pain prescription period was defined between January 8, 2015 and June 30, 2015, and longer-term opioid use examined during 2-year follow-up. Marijuana state policy on July 1, 2015, was assigned: none; medical only; or medical and recreational. PARTICIPANTS: Patients aged 18-62 without cancer diagnosis. MAIN MEASURES: Patient receiving (1) opioid at index; (2) > 7 days' supply of index opioid; (3) opioid during follow-up; and (4) ≥ 90 days' opioid supply during follow-up. Multivariable regression assessed associations between opioid utilization and state marijuana policy, adjusting for age, gender, overall disease burden, mental health treatment, concomitant use of benzodiazepine or muscle relaxant, and previous pain prescription. KEY RESULTS: Of 141,711 patients, 80,955 (57.1%) resided in states with no policy; 56,494 (39.9%) with medical-only; and 4262 (3.0%) with medical and recreational. Patients in states with both policies were more likely to receive an index opioid (aOR = 1.72, 95% CI = 1.61-1.85; aOR = 1.90, 95% CI = 1.77-2.03; P < 0.001) but less likely to receive > 7 days' index supply (aOR = 0.84, 95% CI = 0.77-0.91; aOR = 0.76, 95% CI = 0.70-0.83; P < 0.001) than patients in states with no policy or medical-only, respectively. Those in states with both policies were more likely to receive a follow-up opioid (aOR = 1.87, 95% CI = 1.71-2.05; aOR = 2.20, 95% CI = 2.01-2.42; P < 0.001) than those in states with no policy or medical-only, respectively, and more likely to receive ≥ 90 cumulative follow-up opioid days' supply (aOR = 1.18, 95% CI = 1.07-1.29; P < 0.001) than those in states with no policy. CONCLUSIONS: Our analysis does not support the supposition that access to marijuana lowers use of chronic opioids for pain.

Impact of Healthcare Delivery System Type on Clinical, Utilization, and Cost Outcomes of Patient-Centered Medical Homes: a Systematic Review.

BACKGROUND: As healthcare reimbursement shifts from being volume to value-focused, new delivery models aim to coordinate care and improve quality. The patient-centered medical home (PCMH) model is one such model that aims to deliver coordinated, accessible healthcare to improve outcomes and decrease costs. It is unclear how the types of delivery systems in which PCMHs operate differentially impact outcomes. We aim to describe economic, utilization, quality, clinical, and patient satisfaction outcomes resulting from PCMH interventions operating within integrated delivery and finance systems (IDFS), government systems including Veterans Administration, and non-integrated delivery systems. METHODS: We searched PubMed, the Cochrane Library, and Embase from 2004 to 2017. Observational studies and clinical trials occurring within the USA that met PCMH criteria (as defined by the Agency for Healthcare Research and Quality), addressed ambulatory adults, and reported utilization, economic, clinical, processes and quality of care, or patient satisfaction outcomes. RESULTS: Sixty-four studies were included. Twenty-four percent were within IDFS, 29% were within government systems, and 47% were within non-IDFS. IDFS studies reported decreased emergency department use, primary care use, and cost relative to other systems after PCMH implementation. Government systems reported increased primary care use relative to other systems after PCMH implementation. Clinical outcomes, processes and quality of care, and patient satisfaction were assessed heterogeneously or infrequently. DISCUSSION: Published articles assessing PCMH interventions generally report improved outcomes related to utilization and cost. IDFS and government systems exhibit different outcomes relative to non-integrated systems, demonstrating that different health systems and populations may be particularly sensitive to PCMH interventions. Both the definition of PCMH interventions and outcomes measured are heterogeneous, limiting the ability to perform direct comparisons or meta-analysis.

Value-Based Pharmaceutical Contracts: Value for Whom?

Value-based pharmaceutical contracts (VBPCs) are performance-based reimbursement agreements between healthcare payers and pharmaceutical manufacturers in which the price, amount, or nature of reimbursement is tied to value-based outcomes. VBPCs are often complex, and the nature of who benefits and in what ways can be unclear. We discuss how VBPCs compare with value-based payer-provider arrangements in terms of performance-based reimbursements and alignment of incentives. In addition, we examine how VBPCs can affect costs, clinical outcomes, and access to medications. Because these contracts are unlikely to reduce costs in isolation, we recommend taking a patient-centered approach when developing VBPCs and tying VBPCs to more overarching payer drug cost reduction strategies.

Identifying Outcome Measures for Migraine Value-Based Contracting Using the Delphi Method.

OBJECTIVE: To identify meaningful migraine outcome measures among key stakeholders to inform value-based contracts for migraine medications. BACKGROUND: Value-based contracts linking medication payments to predefined performance metrics aim to promote value through aligned incentives and shared risk between manufacturers and payers. The emergence of new and expensive pharmaceuticals for migraine presents an opportunity for value-based contract development. However, uncertainty remains around which outcomes are most meaningful to all migraine stakeholders. METHODS: This study utilized a Delphi survey to incorporate views from 82 stakeholders, including patients (n = 21), providers (n = 23), payers (n = 10), employers (n = 18), and pharmaceutical company representatives (n = 10). A list of 15 migraine-related outcomes was created from a literature review and subject matter expert consultation. Stakeholders reported on the value of these outcomes through a 5-point Likert scale and selection of their top 3 most meaningful outcomes. All participants except patients and employers also used a 5-point Likert scale to rate the feasibility of collecting each outcome measure. Consensus was defined as ≥75% agreement on the importance and feasibility of an outcome (Likert scores ≥4/5 or selection of an outcome as most meaningful). RESULTS: After 2 rounds, consensus was achieved for importance of 9 outcomes on the Likert scale. "Decrease in migraine frequency" reached 100% agreement (82/82), followed by "increased ability to resume normal activities" (96%, 79/82). When asked to choose the 3 most meaningful outcomes, stakeholders selected "decrease in migraine frequency" (88%, 72/82) followed by "decrease in migraine severity" (80%, 66/82). The 2 measures rated as most feasibly collected were "decrease in emergency department/urgent care visits" (95%, 40/42) and "decrease in migraine frequency" (90%, 38/42). There were statistically significant differences between non-patient and patient stakeholders in selection of "decrease in emergency department/urgent care visits" [20% (12/61) vs 0% (0/21), P = .031]; and employer and patient stakeholders in selection of "decrease in work days missed" [44% (8/18) vs 5% (1/21), P = .006] and "decrease in emergency department/urgent care visits" [22% (4/18) vs 0% (0/21), P = .037] as most meaningful outcomes. CONCLUSIONS: The measures "decrease in migraine frequency" followed by "decrease in migraine severity" were identified as top priority migraine outcome measures.

Payment Reform, Medication Use, and Costs: Can We Afford to Leave Out Drugs?

Medications are one of the fastest growing sources of costs in the health system and the cornerstone of disease management. Despite extensive attention around drug pricing, medications have largely been excluded from CMS-derived, value-based payment models. In this perspective, we synthesize evidence about the impact of three prominent models-primary care-based redesign, ACOs, and bundled payment programs-on medication use, adherence, and costs. We also examine the literature describing similar models implemented by private payors and their relationship with medication use and costs. The exclusion of drug costs from payment reform model design has led to missed opportunities for payors and providers to prioritize effective medication management strategies and has limited our learning about the effects on cost and quality. New CMS-based models are starting to allow greater flexibility in pharmacy benefit design and reward improved medication therapy management. Additionally, health plans, pharmacies, and pharmacy benefit managers are beginning to partner on collaborative value-based pharmacy initiatives. Taken together, these efforts encourage a paradigm shift around drug cost management that more deeply integrates pharmacy into payment and delivery reform with the goal of improving quality and reducing the total cost of care.

Deprescribing in Advanced Illness: Aligning Patient, Clinician, and Health Plan Goals.

Polypharmacy has been linked to adverse outcomes including increased risk of hospitalization, falls, and death and contributes to unnecessary healthcare spending. Deprescribing efforts aim to reduce medication burden while improving or maintaining patients' quality of life. While the practice of deprescribing is gaining momentum, quality measurement and provider reimbursement are barriers that must be addressed for deprescribing to achieve widespread adoption. Because many quality measures are focused on medication use and adherence, deprescribing efforts may negatively impact primary care provider and health plan quality ratings and value-based reimbursement. In addressing this conflict, there are opportunities to proactively align the priorities and incentives of patients, providers, and plans to promote deprescribing. In this report, we propose several actionable steps to address quality and reimbursement-based barriers such as facilitating the exclusion of those engaged in deprescribing efforts from quality measures and the development of deprescribing-based quality measures.

Impact of Star Rating Medication Adherence Measures on Adherence for Targeted and Nontargeted Medications.

BACKGROUND: In 2012, Medicare incorporated medication adherence targeting oral antidiabetic medications, renin-angiotensin system (RAS) antagonists, and statins as highly weighted components in its Star Ratings Program. In the same year, health plans began receiving quality bonus payments for higher star ratings. OBJECTIVE: We aimed to assess how these policy changes affected adherence to targeted and other chronic disease medications in the United States. METHODS: We performed interrupted time series analyses to assess monthly changes in medication adherence from 2010 to 2016 using health plans' Medicare claims submitted to a large pharmacy benefits manager. We conducted 2 sets of analyses. The first examined whether policy changes affected adherence to the 3 targeted therapy classes, and the second assessed the association between policy changes and adherence to 5 chronic disease classes not targeted by star ratings. For the second analysis, we further compared adherence between members who concomitantly used and did not use targeted medications. RESULTS: For star-ratings analyses, we studied 240 811 members on oral antidiabetic medications, 500 958 on RAS antagonists, and 471 135 on statins. Adherence for all star rating-targeted and nontargeted medications increased after 2012 (P < .001). Oral antidiabetic, statin, and RAS antagonist adherence was, respectively, 11.2%, 3.7%, and 8.1% higher than adherence without policy changes (P < .001). Nontargeted antihypertensive and antihyperlipidemic adherence trends were higher among those concomitantly on star rating-targeted medications compared with those who were not (P < .001). CONCLUSIONS: As policy makers strive to identify optimal quality measures for improving healthcare delivery, it is important to consider that incentives can promote improved performance in both targeted measures and related outcomes.

Advancing Access to Cell and Gene Therapies in Medicaid.

This Viewpoint presents the components of the payment model and the implementation challenges among the Center for Medicare and Medicaid Innovation, state Medicaid agencies, patients, and manufacturers.

Waste in the US Health Care System: Estimated Costs and Potential for Savings.

Importance: The United States spends more on health care than any other country, with costs approaching 18% of the gross domestic product (GDP). Prior studies estimated that approximately 30% of health care spending may be considered waste. Despite efforts to reduce overtreatment, improve care, and address overpayment, it is likely that substantial waste in US health care spending remains. Objectives: To estimate current levels of waste in the US health care system in 6 previously developed domains and to report estimates of potential savings for each domain. Evidence: A search of peer-reviewed and "gray" literature from January 2012 to May 2019 focused on the 6 waste domains previously identified by the Institute of Medicine and Berwick and Hackbarth: failure of care delivery, failure of care coordination, overtreatment or low-value care, pricing failure, fraud and abuse, and administrative complexity. For each domain, available estimates of waste-related costs and data from interventions shown to reduce waste-related costs were recorded, converted to annual estimates in 2019 dollars for national populations when necessary, and combined into ranges or summed as appropriate. Findings: The review yielded 71 estimates from 54 unique peer-reviewed publications, government-based reports, and reports from the gray literature. Computations yielded the following estimated ranges of total annual cost of waste: failure of care delivery, $102.4 billion to $165.7 billion; failure of care coordination, $27.2 billion to $78.2 billion; overtreatment or low-value care, $75.7 billion to $101.2 billion; pricing failure, $230.7 billion to $240.5 billion; fraud and abuse, $58.5 billion to $83.9 billion; and administrative complexity, $265.6 billion. The estimated annual savings from measures to eliminate waste were as follows: failure of care delivery, $44.4 billion to $97.3 billion; failure of care coordination, $29.6 billion to $38.2 billion; overtreatment or low-value care, $12.8 billion to $28.6 billion; pricing failure, $81.4 billion to $91.2 billion; and fraud and abuse, $22.8 billion to $30.8 billion. No studies were identified that focused on interventions targeting administrative complexity. The estimated total annual costs of waste were $760 billion to $935 billion and savings from interventions that address waste were $191 billion to $286 billion. Conclusions and Relevance: In this review based on 6 previously identified domains of health care waste, the estimated cost of waste in the US health care system ranged from $760 billion to $935 billion, accounting for approximately 25% of total health care spending, and the projected potential savings from interventions that reduce waste, excluding savings from administrative complexity, ranged from $191 billion to $286 billion, representing a potential 25% reduction in the total cost of waste. Implementation of effective measures to eliminate waste represents an opportunity reduce the continued increases in US health care expenditures.

Avoiding rash decisions about zoster vaccination: insights from cost-effectiveness evidence.

de Boer and colleagues present a cost-effectiveness analysis based in the Netherlands of two vaccines available for the prevention of herpes zoster. Zostavax® was the first vaccine available for the prevention of herpes zoster in older adults. A live-attenuated vaccine, Zostavax is not free of limitations, which include a relatively low efficacy that wanes over time and its contraindication among immunocompromised individuals. The recently available adjuvanted herpes zoster subunit vaccine Shingrix® overcomes some of these limitations. The herpes zoster subunit vaccine is more efficacious than Zostavax, and it can be administered to immunosuppressed individuals. However, the herpes zoster subunit vaccine is considerably costlier and requires a booster injection. In order to clarify the value of each vaccine, de Boer and colleagues compare the cost-effectiveness of no vaccination, and of vaccination with Zostavax or the herpes zoster subunit vaccine in four cohorts of older adults from the perspective of the Netherlands. Whereas neither vaccine was cost-effective under the willingness-to-pay threshold of €20,000 per quality-adjusted life year, the authors find the herpes zoster subunit vaccine to be cost-effective in some scenarios under a €50,000 per quality-adjusted life year threshold.Please see related article: https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-018-1213-5.

Screening for Sexually Transmitted Infections After Cervical Cancer Screening Guideline and Medicaid Policy Changes: A Population-based Analysis.

BACKGROUND: Sexually transmitted infection (STI) screening prevents complications and is cost-effective. Annual screening is recommended in sexually active women below 25 years and older women at increased risk. Cervical cancer (CC) screening guideline changes in 2009 and 2012 recommended less frequent screening, causing concern that STI screening would decrease. Pennsylvania Medicaid implemented a family planning program in 2007 which covered women's health services (including STI testing) for uninsured women. It is unclear how STI screening was affected by these countervailing forces. OBJECTIVE: The main objective of this study was to assess STI screening before and after CC screening guideline changes and family planning program implementation, and to determine factors associated with STI screening. RESEARCH DESIGN: This study was an observational cross-sectional study of Pennsylvania Medicaid administrative claims from 2007 to 2013. SUBJECTS: Sixteen-year-old to 30-year-old women enrolled in Pennsylvania Medicaid. MEASURES: Annual STI screening, defined as receipt of ≥1 STI test in respective 1-year periods. RESULTS: Our population included 1,226,079 women-years for 467,143 women. STI screening increased by 48% between 2007 and 2011, and stabilized by 2013. Odds for STI screening were higher among black compared with white women [adjusted odds ratio (AOR), 2.56; 95% confidence interval (CI), 2.60-3.10]; Hispanic compared with non-Hispanic women (AOR, 1.42; 95% CI, 1.39-1.46); family planning program enrollees (AOR, 1.42; 95% CI, 1.40-1.45); and urban compared with rural residents (AOR, 1.05; 95% CI, 1.03-1.06). CONCLUSIONS: STI screening dramatically increased between 2007 and 2011. Potential reasons are family planning program implementation, increased urine/vaginal testing, and reporting improvements. It is reassuring that STI screening did not decrease despite CC screening guideline changes. Between 2011 and 2013, rates stabilized at 45% among all women and 60% among sexually active women below 25 years, suggesting opportunities for improvement.

Impact of community pharmacist-provided preventive services on clinical, utilization, and economic outcomes: An umbrella review.

Preventable diseases and late diagnosis of disease impose great clinical and economic burden for health care systems, especially in the current juncture of rising medical expenditures. Under these circumstances, community pharmacies have been identified as accessible venues to receive preventive services. This umbrella review aims to examine existing evidence on the impact of community pharmacist-provided preventive services on clinical, utilization, and economic outcomes in the United States (US). We included systematic reviews, narrative reviews and meta-analyses published in English between January 2007 and October 2017. Of 2742 references identified by our search strategy, a total of 13 research syntheses met our inclusion criteria. Included reviews showed that community pharmacists are effective at increasing immunization rates, supporting smoking cessation, managing hormonal contraception therapies, and identifying patients at high risk for certain diseases. Moreover, evidence suggests that community pharmacies are especially well-positioned for the provision of preventive services due to their convenient location and extended hours of operation. There is general agreement on the positive impact of community pharmacists in increasing access to preventive health, particularly among patients who otherwise would not be reached by other healthcare providers. The provision of preventive services at US community pharmacies is feasible and effective, and has potential for improving patient outcomes and health system efficiency. However, high-quality evidence is still lacking. As the healthcare landscape shifts towards a value-based framework, it will be important to conduct robust studies that further evaluate the impact of community pharmacist-provided preventive services on utilization and economic outcomes.

Prenatal and Postpartum Care Disparities in a Large Medicaid Program.

Objectives Pennsylvania's maternal mortality, infant mortality, and preterm birth rates rank 24th, 35th, and 25th in the country, and are higher among racial and ethnic minorities. Provision of prenatal and postpartum care represents one way to improve these outcomes. We assessed the extent of disparities in the provision and timeliness of prenatal and postpartum care for women enrolled in Pennsylvania Medicaid. Methods We performed a cross-sectional evaluation of representative samples of women who delivered live births from November 2011 to 2015. Our outcomes were three binary effectiveness-of-care measures: prenatal care timeliness, frequency of prenatal care, and postpartum care timeliness. Pennsylvania's Managed Care Organizations (MCOs) were required to submit these outcomes to the state after reviewing administrative and medical records through a standardized, validated sampling process. We assessed for differences in outcomes by race, ethnicity, region, year, and MCO using logistic regression. Results We analyzed data for 12,228 women who were 49% White, 31% Black/African American, 4% Asian, and 15% Hispanic/Latina. Compared to Black/African American women, white and Asian women had higher odds of prenatal and postpartum care. Hispanic/Latina women had higher frequency of prenatal care than non-Hispanic women. Pennsylvania's Southeast had lower prenatal care and Northwest had lower postpartum care than other regions. Prenatal care significantly decreased in 2014 and increased in 2015. We observed differences between MCOs, and as MCO performance diminished, racial disparities within each plan widened. We explored hypotheses for observed disparities in secondary analyses. Conclusions for Practice Our data demonstrate that interventions should address disparities by race, region, and MCO in equity-promoting measures.

Cultural implications of mentoring in sub-Saharan Africa: a qualitative study.

CONTEXT: Although many studies have demonstrated the benefits of mentoring in academic medicine, conceptual understanding has been limited to studies performed in North America and Europe. An ecological model of mentoring in academic medicine can provide structure for a broader understanding of the role of culture in mentoring. OBJECTIVE: The goal of this study was to explore the role of culture in the development and maintenance of mentoring relationships within the context of the University of Malawi College of Medicine. METHODS: A qualitative study using in-depth, semi-structured interviews and thematic analysis was conducted to explore the meaning of mentorship at the study institution. Criterion sampling was used to identify and recruit medical students, interns, registrars and faculty members. Study team members developed a codebook through open coding and applied it to all interview transcripts. Thematic analysis was used to identify and categorise themes according to an ecological model. RESULTS: A total of 46 participants from two major centres in Malawi were interviewed. Themes were identified within three domains: the intrapersonal; the interpersonal, and the institutional. Intrapersonal themes included Malawian politeness, mentoring needs, and friendliness and willingness to help. Interpersonal themes included understanding the role of the mentor, respect for elders, personal and professional boundaries, and perceptions of others. Institutional themes included the supervisor versus mentor, time pressures, tension about the scope of training, and the mentoring cycle. CONCLUSIONS: This study highlights the strengths of and challenges imposed by culture to the provision of mentoring relationships at the study institution. It also highlights the central role of culture in mentoring and proposes an updated model for mentoring in academic medicine. This model can inform future research on mentoring and may serve as a model in the larger effort to provide faculty development in mentoring across sub-Saharan Africa.

Specialization training in Malawi: a qualitative study on the perspectives of medical students graduating from the University of Malawi College of Medicine.

BACKGROUND: There is a critical shortage of healthcare workers in sub-Saharan Africa, and Malawi has one of the lowest physician densities in the region. One of the reasons for this shortage is inadequate retention of medical school graduates, partly due to the desire for specialization training. The University of Malawi College of Medicine has developed specialty training programs, but medical school graduates continue to report a desire to leave the country for specialization training. To understand this desire, we studied medical students' perspectives on specialization training in Malawi. METHODS: We conducted semi-structured interviews of medical students in the final year of their degree program. We developed an interview guide through an iterative process, and recorded and transcribed all interviews for analysis. Two independent coders coded the manuscripts and assessed inter-coder reliability, and the authors used an "editing approach" to qualitative analysis to identify and categorize themes relating to the research aim. The University of Pittsburgh Institutional Review Board and the University of Malawi College of Medicine Research and Ethics Committee approved this study and authors obtained written informed consent from all participants. RESULTS: We interviewed 21 medical students. All students reported a desire for specialization training, with 12 (57%) students interested in specialties not currently offered in Malawi. Students discussed reasons for pursuing specialization training, impressions of specialization training in Malawi, reasons for staying or leaving Malawi to pursue specialization training and recommendations to improve training. CONCLUSIONS: Graduating medical students in Malawi have mixed views of specialization training in their own country and still desire to leave Malawi to pursue further training. Training institutions in sub-Saharan Africa need to understand the needs of the country's healthcare workforce and the needs of their graduating medical students to be able to match opportunities and retain graduating students.