Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:

Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_612_Библиотеки_им_академика_М_И_Перельмана

.pdf
Скачиваний:
0
Добавлен:
30.08.2026
Размер:
67 Мб
Скачать
For those starting in research or expanding into new areas of inquiry, this rigorous question development process highlights the value of a research team—experience with other methods, perspectives on ethics and feasibility, and an overarching focus on clinical and research alignment.
SELECTING THE BEST DESIGN
Just as the surgeon trains to master options for reconstruction, the researcher builds a skillset of research methodologies. For a specific research idea, there are several methods that can be applied with tailoring of the research question. Other research proposals are best achieved with a specific method. For example, exploring the mediators of inequities in access to breast reconstruction could be done with interviews of patients considering reconstruction (qualitative), analysis after change to policy (causal inference), and with comparison between groups (traditional quantitative approaches). The choice among methods requires matching the research team’s skill (established or evolving) with the aspect of the problem that is most urgent. Similarly, if a researcher starts reading a study on a topic and it does not use the design they would have used, they should review the choices the team made as well as the alternative aspects of the problem their envisioned design would have elucidated.
Once familiar with a methodology, a surgeon-scientist can apply it to other research questions. It is possible to work in reverse and try to design a project using a specific methodology, but we recommend early researchers keep their options open to find high-impact research opportunities and select the design for the question rather than vice versa. This is especially powerful because it encourages a researcher to keep a running list of possible research questions that can be revisited and reviewed for feasibility across the course of a career. Even for a surgeon without plans to conduct research, working knowledge of a method can enable careful consideration of published results. A busy clinical surgeon familiar with the principles of research design can vet the new evidence around a new technique before changing practice. In the same way peer-reviewed
https://t.me/med1917
publications and presentations can disseminate changes to practice; they can introduce a researcher to a new methodology that facilitates addressing a problem from a different angle. In this way, even an experienced researcher can use a new project to master another method, as part of a collaborative team.
Qualitative Methods
Quantitative methods are the most common in medical and surgical research. However, researchers must consider all methodologies to explore novel questions. Qualitative research is underused, but it can also be combined with quantitative data through mixed methods approaches.37 For example, qualitative and quantitative methods have both been successfully applied to a range of clinical problems in medicine and plastic surgery. Qualitative approaches yield rich, in­depth stakeholder experiences. However, surgical researchers and reviewers are often most comfortable with quantitative approaches.
Although qualitative undertakings can be time consuming, it is important to recognize their strengths. There is increased attention to developed patient-reported outcome measures (eg, BREAST-Q
32
and WOUND-Q38), and clinicians and researchers are starting to prospectively collect these measures. Although these are now available as scales, they were developed through interviews, pilot groups, and psychometric validation. Lane and colleagues utilized the BREAST-Q, BODY-Q, and patient health questionnaire-9 to investigate patient satisfaction and quality of life after gender­affirming mastectomy. These patient-reported outcome measures allowed for assessing the resultant improvement of body image, chest appearance, and mood.32 In the absence of these established tools, or if a researcher wanted to further understand the perioperative experience for these patients, a qualitative approach to explore the emotions around this area could be considered with active listening or semi-structured interviews.
Quantitative Methods
Cohort
https://t.me/med1917
Cohort studies can be either prospective or retrospective in their data collection. Traditional data sources at the national level are the National Surgical Quality Improvement Program (NSQIP) and the Tracking Operations & Outcomes for Plastic Surgeons (TOPS). Although each is maintained by a national surgical society, TOPS (the American Society of Plastic Surgeons) contains variables of specific interest to plastic surgeons that are not captured in NSQIP (the American College of Surgeons).39-41 These data sources should be considered early, as an inherent need for patient, surgeon, facility, or community data that are not captured in these datasets may require the design of a new prospective dataset. Cohort studies are traditionally longitudinal, in that a subject is enrolled or monitored over time. Beyond the limitations of data collected retrospectively, this can be challenged by loss to follow-up. Additionally, if an outcome is rare or not clearly coded in these datasets, it often cannot be captured sufficiently through preexisting data. Although this realization is often the start of multi-institutional research collaboratives, not every researcher may be positioned to launch such an endeavor at the time of reaching that conclusion. A cohort study that is prospective, or collecting data as patients are enrolled, can capture detailed data of interest to researchers but requires a time-intensive and costly research infrastructure.
Case-Control
Case-control designs are best suited for rare outcomes. In epidemiological research, this traditionally starts with rare cancers that may develop decades after exposure. This then identifies the cases, who are then compared with controls without the outcome. These results are presented as odds ratios with descriptive summary statistics. For example, researchers interested in the surgical site infection risk of immediate implant-based breast reconstruction reviewed 10 years of cases to identify all cases with postoperative surgical site infections and match them to control cases. For a rare complication, this method allowed for identification of obesity, hypertension, and neoadjuvant therapy as risk factors. When done to explore exposures not captured in real time in the electronic medical
https://t.me/med1917
record, this approach is subject to recall bias. That is, patients who experienced a complication may remember more details of their postoperative course than patients who experienced an uneventful recovery.
Cross-Sectional
For questions of prevalence or volume, cross-sectional study designs can be useful. These are also among the most straightforward designs for analysis. There are several examples in the literature, ranging from measurement of the imposter phenomenon in academic plastic surgery to the financial toxicity among female patients with breast cancer undergoing reconstruction.
42,43
Although these descriptive studies can be valuable, they should be designed to identify targets for further investigation. Consider, for example, the many descriptive studies of inequitable outcomes. These cross-sectional studies are limited in that they cannot infer causality, but they can benchmark outcomes or suggest prevalence of the problem. Similarly, many emerging studies of price transparency and CMS payment systems are cross-sectional in design.
11,44
Randomized Controlled Trial
These trials are the standard in biomedical research. Although perhaps most often associated with drug development and testing, these trials have been implemented to evaluate causality of technical and systemic changes. For instance, the hernia research group at the Cleveland Clinic designed a randomized controlled trial (RCT) to investigate the role of transfascial fixation sutures in large ventral hernia repairs. Patients who met inclusion criteria were randomized by a member of the research team intraoperatively. Patients received either 8 transfascial sutures or 0 transfascial sutures (with 8 sham incisions).45 Although this study demonstrates use of a randomized trial to investigate a technical approach to surgery with a traditional placebo arm, it also highlights the challenges of conducting them. This research group was unique in that it was already in place with a prospective data collection infrastructure. Conducting an RCT,
https://t.me/med1917
whether at one institution or across multiple institutions (known as a multi-institutional trial), requires significant investment of time and resources as well as buy-in from those who are being told their decisions are being determined by a random allocation. This was illustrated with the National Institutes of Health–funded Wrist and Radius Injury Surgical Trial (WRIST), in which patients with unstable distal radius fractures were randomized to percutaneous pinning, external fixation, or internal fixation.
46
Furthermore, the upfront design should consider the unmeasured features of the institutions or surgeons. For example, the 0 or 8 suture approaches of the hernia group at the Cleveland Clinic may not reflect the number of fixating sutures used by other hernia surgeons. In this choice of intervention (no more than 8 sutures), the trial design limited generalizability to other surgeons who use more fixating suture. In contrast, the approach in WRIST includes a range of surgical approaches to unstable distal radius fracture. Similarly, exclusion criteria can also limit generalizability to other patients. A trial that excludes patients with prior radiation will not answer the question of the best approach for those patients.
By using randomization that balances both measured and unmeasured confounders, RCT results infer causality. Although most often considered to imply randomization at the level of the patient, RCTs can randomize treatment at the patient, surgeon, hospital, or neighborhood level. This can be accomplished with cluster randomization. One excellent application of a cluster randomized trial divided the master list of abandoned houses in Philadelphia into 63 clusters. Each cluster of houses was then randomized between three study arms: full remediation, trash cleanup only, and no intervention to evaluate gun violence and subjective safety. There was a decrease in gun violations and gun assaults in the full remediation arm.47 Similarly to geographic clusters, clustered randomization can occur at the hospital or surgeon level.
RCTs are often performed in ideal conditions; that is, a drug or treatment may be provided at no cost or participants’ travel may be covered. Researchers interested in the “real-world” impact of an intervention may also consider the pragmatic trial. Key principles of a
https://t.me/med1917
pragmatic RCT are enrolling population close to practice population, community sites rather than research sites, appropriate comparison arm (standard of care rather than placebo), and relevant patient­reported outcomes and clinical outcomes.
48
Causal Inference
When randomization is not feasible, there is growing attention to quasi-experimental designs to consider causality from observational data.49 As the healthcare system continues to become more complex, controlling allocation to receipt of a surgical technique, coverage for a reconstructive procedure, or postoperative resources is not always feasible. A researcher cannot petition to reallocate states to Medicaid expansion to study the causal relationship between this policy and reconstructive outcomes.50 However, researchers can use the “natural experiment” of expansion to compare rates of breast cancer screening and reconstruction between the states with and without Medicaid expansion.
15,18,51
These study designs are being introduced to surgical research from the field of econometrics.
49
Selecting the Method
Facing complex clinical and health system problems, it can be overwhelming to consider every possible question-method pairing. To decide on which question-method will yield high-impact research, we first consider what is most important to us and the field to answer. Next, we consider which of those question-method pairings is the most feasible. For instance, given the lack of all-payer database, it is not currently possible to assess the national outcomes of uninsured hand trauma patients. However, ensuring those patients have access to reconstructive surgery requires a viable safety net. This further underscores the importance of creating a conceptual model early. Even if one question or approach is not yet feasible, there are often other angles. A researcher must consider what aspect of the identified problem most bothers them to find a motivating and high­impact question.
https://t.me/med1917
ANALYTIC PLAN
The best analytic plans are based on conceptual models. Even when designing an analysis of a retrospective dataset, the researcher should understand the potential for interactions, the role of mediators, and possible confounders. For some projects the researcher is designing an analysis with statistical experts, but especially early on the researcher or research fellow is also performing the analysis. At this point, with a clearly defined research question and study design, the conceptual model is more detailed than the models created earlier in the process. It now may include specific variables and their relationship to desired primary and secondary outcomes.
A conceptual model is often sketched out with pen and paper and redrawn multiple times. This is because additional relationships will emerge as a literature review proceeds. It can also allow for visual representation of potential confounders or other factors linked to the independent variable that may enhance, weaken, or reverse the association between the independent and dependent variables. A confounder must meet three criteria: it must cause the outcome, be associated with the independent variable, and not be the connection between the two. One of the most common forms of confounding in surgical outcomes research is confounding by indication. This is seen when the indication for a given treatment can also impact the outcome.52 For instance, severe tissue loss may require more intensive reconstruction with longer length of stay. The degree of tissue loss not only leads to the choice of intervention (reconstruction) but is also linked to longer lengths of stay regardless of intervention. Of the study designs, RCTs are often suggested to control for measured (eg, tissue loss extent) and unmeasured (eg, lack of family support at home) confounders. However, researchers will also come across statistical methods including propensity scores and logistic regression to adjust for these confounders. Although all methods have limitations, the investigator must recognize potential confounders and discuss them thoroughly.
Although much of the planning process can feel broad, the analytic plan should be granular in detail. The details of this plan will differ
https://t.me/med1917
based on the chosen study design. For cross-sectional studies, descriptive statistics are sufficient. For a national study, multilevel modeling approaches may be necessary. Mixed methods approaches can vary in combining quantitative and qualitative findings or proceeding sequentially. However, we recommend a standard analytic plan format that anticipates the key elements necessary to initiate the project.
Population
Research questions are often written with a broad description of the population of interest. The analytic plan narrows that broad group into a series of criteria. For example, instead of “women with breast cancer,” a study population might be patients aged 18 to 65 years referred for immediate reconstruction evaluation at a tertiary academic center. The population for this breast cancer reconstruction access project could range from the residents of a community to a national comparison across cancer centers. This granularity asks the researcher to define the population, the sampling method, and the inclusion/exclusion criteria. In other words, what designation will be used to determine cancer center status? Are centers excluded if the designation was only granted for a portion of the years of the study? This step also requires recognizing the predetermined inclusion or exclusion criteria of existing datasets.
Data Source
Existing data sources often determine population for retrospective work. That is, Medicare claims data narrow the patient population to Medicare-eligible patients receiving care from plastic surgeons participating in Medicare for procedures reimbursed by Medicare. If the desired study population is those 18 to 65 years old and only Medicare claims are available, the project is not feasible. However, it could also direct the team to explore options for commercial insurance claims databases. Many prospective studies are launched to fill gaps in existing data sources. That is, if the population of interest is not well represented for the outcome or procedure of interest, a larger prospective study is indicated.
https://t.me/med1917
If a dataset is being created, whether collected prospectively or combined from existing databases, this data source section should include the proposed data dictionary. This includes variable names (either already assigned or what they will be named), format of variables, units of measurement, and any codes (eg, 1 = yes, 0 = no). If datasets are being downloaded from government sources (eg, CMS), record the web address and description of the dataset for future reference. The researcher should also download complete data dictionaries for preexisting databases being used. It is common for updated data to be uploaded between the time of analysis and publication, and a project should cite the exact dataset that was used.
For qualitative approaches, the data source section includes the preliminary interview guide or survey. A rigorously developed semi­structured interview guide is revised iteratively. Similarly, a survey is piloted and should be validated before wide distribution. Most institutional review boards (IRBs) will expect these research tools to be submitted for research project approval. The research team should work with the IRB to follow requirements for revisions and modifications to planned research tools.
CONSIDERING LIMITATIONS AND BIAS
Many methodological strengths and critiques will become more apparent once the analytic plan is designed. These may not be issues with feasibility, but effective research projects consider them early. In other words, a researcher should begin thinking about possible reviewer critiques of the plan before the data collection is underway. Potential limitations and bias should be considered and discussed during the design process. The presence of limitations or the recognition of bias should not stop a study from proceeding. Every study design carries its own limitations. However, early recognition of these limitations can allow for early pivots in study design and produce stronger final manuscripts.
There are situations when these limitations will prevent the research team from answering the study question. For example, a proposed project to investigate the experience of patients whose
https://t.me/med1917
primary language is not English at an institution that rarely provides care to those patients is difficult. Identifying limitations early can facilitate improvement to study design and help find another site with which to collaborate. For example, a project designed to investigate barriers to breast reconstruction with proposed recruitment of patients at a plastic surgery clinic carries a selection bias. Recognizing this prior to the start of the project provides opportunity to overcome this bias by recruiting from a breast oncology clinic or even mobile mammography unit. Rather than only considering these limitations when prompted by a reviewer, a researcher who proactively anticipates and mitigates these critiques will lead a more rigorous study.
There is no design free of bias, but rigorous studies should clearly identify potential sources of bias. The internal validity of a study, its ability to produce valid results, is threatened by random error (chance) and systematic error (bias). This is a systematic variation that causes the study’s estimated association (eg, risk ratio) to differ from the true association. This contrasts with chance. There are three categories of systematic bias: selection bias, confounding and information bias.53 Confounders have been discussed previously. Selection bias is most well-known as it is commonly discussed in study enrollment. This can occur due to preexisting referral patterns that may direct more complex cases to a study site (referral bias) or patients who decline to participate in a study due to lack of reliable transportation (nonrespondent bias).54 Information bias captures measurement error leading to misclassification of exposures and outcomes, interviewer bias, and recall bias.
TAKING THE SO WHAT? TO NOW WHAT?
A well-crafted research plan should include the anticipated implications of findings. Similarly, the manuscript can be partially drafted even before the analysis is complete. However, an impactful research project should not end with publication. Findings published in high-impact journals (as scored by impact factor) are well regarded for the audience they could reach. However, dissemination of finding can include sharing results not only among professional
https://t.me/med1917