SEARCH

SEARCH BY CITATION

Keywords:

  • colorectal neoplasms;
  • early detection of cancer;
  • costs and cost analysis;
  • patient navigation;
  • intervention studies

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES

BACKGROUND

Colorectal cancer (CRC) screening is cost-effective but underused. The objective of this study was to determine the cost-effectiveness of a mailed standard intervention (SI) and tailored navigation interventions (TNIs) to increase CRC screening use in the context of a randomized trial among primary care patients.

METHODS

Participants (n = 945) were randomized either to a usual care control group (n = 317), to an SI group (n = 316), or to a TNI group (n = 312). The SI group was sent both colonoscopy instructions and stool blood tests irrespective of baseline preference. TNI group participants were sent instructions for scheduling a colonoscopy, a stool blood test, or both based on their test preference, as determined at baseline; then, they received a navigation telephone call. Activity cost estimation was used to determine the cost of each intervention and to compute incremental cost-effectiveness ratios. Statistical uncertainty within the base case was assessed with 95% confidence intervals derived from net benefit regression analysis. The effects of uncertain parameters, such as the cost of planning, training, and involvement of those receiving “investigator salaries,” were assessed with sensitivity analyses.

RESULTS

Program costs of the SI were $167 per participant. The average cost of the TNI was $289 per participant.

CONCLUSIONS

The TNI was more effective than the SI but substantially increased the cost per additional individual screened. Decision-makers need to consider cost structure, level of planning, and training required to implement these 2 intervention strategies and their willingness to pay for additional individuals screened to determine whether a tailored navigation would be justified and feasible. Cancer 2014;120:1042–1049. © 2013 American Cancer Society.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES

Estimates indicate that there will be 142,820 new cases and 50,830 deaths from colorectal cancer (CRC) in 2013.[1] CRC screening can detect colorectal adenoma, a precursor to CRC. The American Cancer Society (ACS) and the United States Preventive Services Task Force (USPSTF) encourage screening in adults who are aged ≥50 years, asymptomatic, and at average risk for CRC.[2]

Colonoscopy every 10 years and annual stool blood testing (SBT) are the most frequently performed CRC screening strategies.[3] Although US rates of CRC screening are increasing, they lag behind those for breast and cervical cancer screening.[4] Healthy People 2020 called for greater than 70% CRC screening rates.[5] To achieve sustainable high CRC screening rates, primary care providers need to implement cost-effective behavioral interventions. Whereas mailed contacts and reminders have a modest impact on CRC screening rates,[6] personally tailored patient programs may achieve greater gains.

Few studies have evaluated the cost-effectiveness of CRC screening promotion strategies.[7, 8] Estimates for CRC promotion range from $11 to $978 per additional individual screened.[9, 10] The former estimate was for a low-cost intervention directed toward disadvantaged populations in primary care clinics, whereas the higher cost was for a sophisticated tracking system in the Veterans Administration Health System. Elsewhere, we reported on the cost-effectiveness of targeted and tailored behavioral outreach interventions to increase CRC screening within an urban family practice in Philadelphia. Our estimates of the mean intervention cost in that project ranged from $42 to $200.11

Here, we present findings from a randomized trial designed to test the cost-effectiveness of a standard intervention (SI) and a tailored navigation intervention (TNI) on CRC screening use among patients in the Christiana Care Health System (CCHS). CCHS provides health care to approximately 66% of the Delaware population. The system includes 2 acute care hospitals with over 42,000 inpatient admissions and more than 125,000 annual emergency department visits and a network of primary care practices in family medicine, internal medicine, pediatrics, and obstetrics/gynecology. We recently reported the research design and screening outcomes of SI and TNI compared with usual care (control).[12]

Interventions that help patients address barriers to screening and help providers deliver the screening message increase CRC screening adherence.[13-17] The population outreach strategies used in this randomized trial and the detailed screening outcome and cost data provide information for decision-makers' assessment of CRC prevention methods.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES

We conducted a prospective, 3-group, randomized controlled trial in 10 primary care practices affiliated with CCHS between 2007 and 2011.12 Computerized and manual reviews of electronic medical records (EMRs) identified potentially eligible patients ages 50 to 79 years with no prior diagnosis of colorectal neoplasia or inflammatory bowel disease who had complete contact information, were noncompliant with USPSTF CRC screening guidelines, and had visited the participating practices within the previous 2 years.[18] Visit criteria were selected to increase the availability of current contact information. A manual review of the EMR excluded patients who were up to date with screening. Potentially eligible patients were verified and consented by telephone after receiving an introductory letter. Thomas Jefferson University and the CCHS institutional review boards approved the study (registered clinical trial NCT00617071).

Intervention

After the baseline telephone survey, participants were randomly assigned to 1 of 3 study groups. The control group received usual care. The SI group received a booklet on CRC screening, a personalized letter with a colonoscopy information number, and an SBT kit. A nurse gave callers the telephone number of a primary care practice-approved gastroenterology or surgical provider. Participants were mailed a reminder letter 30 days after randomization.

TNI participants were mailed the CRC screening booklet and materials keyed to their baseline survey-determined preferred CRC screening test. Those who preferred colonoscopy received a letter and a message page with the colonoscopy information number. Participants with an equal preference for colonoscopy and SBT received a letter and message page with the colonoscopy information number and an SBT kit. Participants who preferred SBT received a letter and a message page with an SBT kit.

Approximately 10 days after the TNI mailing, a nurse navigator initiated a structured navigation call. The navigator confirmed receipt of the mailing and asked the participant whether they had any questions about the materials. Materials were resent if the respondent reported that they were not received. The navigator reassessed the participants' screening test preference, encouraged completion of the test by providing instructions as appropriate for completing the SBT, identifying sites available for colonoscopy, and encouraging participants to schedule colonoscopies. If the participant changed their baseline test preference, then additional materials were mailed within 4 days. A reminder mailing with an SBT kit was sent at 30 days for the TNI group participants who preferred SBT or had an equal preference for SBT and colonoscopy screening. For those who preferred colonoscopy, a reminder mailing was sent at 90 days. In the SI group, 316 participants were mailed an initial SBT, and 261 (83%) received a second mailed SBT. In the TNI group, 177 participants received an initial mailed SBT, and 74 (42%) received a second mailed SBT.

Effect Estimation

Screening results were combined with an estimate of intervention costs to assess the cost-effectiveness of the intervention strategies, including a secondary analysis of TNI subgroups compared with controls according to whether participants expressed a preference for a screening modality.[12] Endpoint chart audits were conducted 6 months and 12 months after initial random assignment to allow for delays because of scheduling and colonoscopy completion. Adherence was defined as any ACS-recommended or USPSTF-recommended CRC screening test.[2, 3, 19-22] The literature indicates that there is modest over-reporting of CRC screening in survey responses[23] and modest under-reporting in EMRs.[24] No single source of CRC screening performance is error-free, and screening from 1 source may not be captured by another. Therefore, any screening reported on the endpoint survey that was identified in the laboratory records or in the medical records was counted as long as the report was accompanied by a date within the target window.

Cost Estimation

Labor, supply, and overhead costs were estimated from the provider perspective in 2011 US dollars. Data sources included questioning staff to determine the approximate time required to complete tasks, study invoices, and current market prices for supplies. Total costs for each activity were divided among the intervention arms according to the number of patients in each group and the number of patients involved in each activity.

Baseline survey costs were estimated at 33% of recorded time, eliminating the time required for research and including the time to introduce the study, assess eligibility, and ask questions about test preference. If baseline survey time was missing, then 33% of the average was applied. Survey costs were included only for the TNI group, which required a preference-based mailing. Participants in the SI group received a generic mailing that could be sent to members of a practice without an advance survey.

Personnel cost

Personnel costs were computed by multiplying the hours spent on each activity by adjusted salary per hour, using the base salary for each position to calculate the adjusted salary per hour.[25] Task times were estimated retrospectively. Training and planning meeting times were adjusted by allocating training and planning costs by the percentage of staff member time required within each study arm.

Supply cost

Supplies included questionnaires, printing costs, envelopes, paper, postage, cell phone use, and other resources used to enroll participants and to produce intervention materials. Costs were calculated by quantifying the materials for each activity and pricing according to current market rates. Professionally printed materials were accounted from project invoices.

Overhead

Administrative services, utilities, computer use, software, and office space were estimated by multiplying the total direct cost (base) by a hypothetical indirect rate of 30%, consistent with the health care costing literature.[26] The indirect cost base is the summed personnel and supply cost.

Cost-Effectiveness Analysis

Incremental cost-effectiveness ratios (ICERs) represented the cost per additional individual screened, comparing the control group first with the SI group and then with the TNI group. Investigator time and salary were examined with sensitivity analysis. The base case scenario applied a 50% estimate of investigator time (range, 25%-75%), assuming different proportions of time spent on different research aspects. Investigator salaries were adjusted to 2 practice managers and 1 primary care physician levels.[27] Mid-level salaries were used for the base case, with a range from 15% higher to 15% lower. We examined finer gradations over the salary range, demonstrating differences only between $1 and $2 from the base case per individual cost.

Statistical uncertainty was assessed with net benefit regression analysis and cost-effectiveness acceptability curves (CEACs), which indicate the probability that an intervention is cost-effective relative to an alternative for a range of willingness-to-pay values.[28, 29] We computed the net benefit for the individual as follows: NBi = λ Ei − Ci, where λ represents the decision-maker's willingness to pay for a unit of E, and C is the cost of providing the intervention to person i. NBi is the dependent variable in the net benefit regression model NBi = β0 + β1I + ε, where I is the indicator for the intervention received, which is set to 1 if the participant received SI and 0 if the participant was in the control group. The coefficient β1 measures the mean difference in net benefit between the SI and control groups. The SI is cost-effective relative to control if β1 is positive. The 1-sided P value for β1 determines the probability that SI is cost-effective relative to control, which forms 1 point on the x-axis in the cost-effectiveness acceptability (Fig. 1).[28] The net benefit regression was estimated for several hypothetical λ values, and the results were used to determine the remaining points on the CEAC.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES

The baseline survey was completed by 945 respondents in the control group (n = 317), the SI group (n = 316), and the TNI group (n = 312). The majority of respondents were women, white, aged <60 years, married, and had more than a high school education (Table 1). Note that occasional imbalances of some baseline characteristics do not normally indicate a flaw in the randomization process, and differences in baseline characteristics are a result of chance and not bias.[30] The implementation of randomization procedures was closely monitored to avoid any flaws in randomization.

Table 1. Participants' Baseline Characteristics
 No. of Participants (%)a
CharacteristicUC Group, N = 317SI Group, N = 316TNI Group, N = 312
  1. Abbreviations: SI, standard intervention; TNI, tailored navigation intervention; UC, usual care.

  2. a

    Counts may not sum to each group's total because of occasional missing data.

  3. b

    Nonwhites were primarily African Americans but also included some Asians and members of other racial groups.

Age, y   
50-59223 (70)222 (70)213 (68)
60-6969 (22)70 (21)67 (21)
70-7925 (8)24 (10)32 (10)
Sex   
Men181 (57)201 (64)207 (66)
Women136 (43)115 (36)105 (34)
Race   
White248 (78)249 (79)243 (78)
Nonwhiteb69 (22)67 (21)69 (22)
Education   
≤High school131 (42)136 (43)134 (43)
≥High school183 (58)177 (57)175 (57)
Marital status   
Married or living as married189 (60)200 (63)197 (64)
Single/divorced/widowed127 (40)116 (37)113 (36)
Preferred screening test   
Stool blood test58 (18)75 (24)52 (17)
Equal preference for stool blood test and colonoscopy151 (48)119 (38)125 (40)
Colonoscopy108 (34)122 (39)135 (43)

Intervention Costs

The base-case cost per eligible patient for the SI group and TNI group was $167 dollars and $289 per individual, respectively (Table 2). Within the SI study arm, approximately 79% of the cost was for personnel, and 21% was for supplies. The TNI study arm cost was approximately 88% personnel, and only 12% was for supplies. The TNI group added time for the completion of a portion of the baseline survey and for the production of tailored mailings, training and time required for completing navigation calls, and time for any follow-up to address patient barriers. The TNI group incurred less supply cost because the SBT kits were not mailed to all participants. For both study arms, approximately 30% of the total cost was for the recruitment of eligible members of the target population.

Table 2. Cost of the Interventions
 Cost, $
ActivityUC (n = 317)SI (n = 316)TNI (n = 312)
  1. Abbreviations: NA, not applicable; SI, standard intervention; TNI, tailored navigation intervention; UC, usual care.

  2. a

    Overhead was estimated at 30% of the direct cost.

Identify target population026,463.7826,128.80
Training and planning meetings03846.2721,823.44
Brochure development02426.362395.64
Baseline survey0NA1245.73
SI intervention mailing04922.74NA
SI reminder mailing02892.37NA
TNI intervention mailing0NA5550.46
TNI intervention call0NA8949.54
TNI reminder0NA3263.24
Total direct cost040,551.5269,356.85
Overhead a012,165.4620,807.06
Total cost052,716.9890,163.91
Cost per individual0166.83288.99

Cost-Effectiveness

Table 3 presents estimates of the cost-effectiveness analysis moving from a less resource-intensive SI to a more resource-intensive TNI. Because the purpose of the economic evaluation was estimation and not to test a hypothesis, we used the point estimates as the “best” available estimates of program effects and costs.[29]

Table 3. Incremental Cost-Effectiveness: Intervention Cost per Additional Individual Screened
InterventionCost, $aIncremental Cost, $Effect: Percentage Screened (OR)Incremental Effectiveness, %ICER [95% CI], $
  1. Abbreviations: CI, confidence interval; ICER, incremental cost-effectiveness ratio, OR, odds ratio; SI, standard intervention; TNI, tailored navigation intervention; UC, usual care.

  2. a

    Costs estimated with the net benefit regression model yield slightly different values than the average costs calculated in Table 2.

UC017.98 (1.00)
SI166.83166.8336.39 (2.68)18.41906.19 [665.83-1437.08]
TNI288.99122.1642.63 (3.48)6.241957.69 [880.01-9043.28]

Approximately 18% of participants in the control group (usual care) were screened compared with 36.4% of the SI group and 42.6% of the TNI group at 12 months. The cost per additional individual screened (ICER) was $906 (95% confidence interval [CI], $666-$1437) comparing the control group with the SI group. When we compared the SI group with the TNI group, the ICER was $1958 (95% CI, $880-$9043). The results were sensitive to the range of investigator time and salaries when comparing the TNI with the SI, but not when comparing the SI versus the control intervention (Table 4). The ICER for the SI compared with the control intervention increased only 2.8% from the low-value combination of cost to the high-cost combination of investigator salary and time. When comparing the SI with the TNI, the ICER increased 25.7% using the higher values. Even using the low estimate, the cost per additional individual screened was $1758 when comparing the SI with the TNI.

Table 4. Effect of One-Way and Two-Way Sensitivity Analyses on Incremental Costs and Incremental Cost-Effectiveness Ratiosa
  Incremental Cost, $ICER, $
Cost Scenario AdjustmentsRange of ValuesSI vs UCTNI vs SISI vs UCTNI vs SI
  1. Abbreviations: ICER, incremental cost-effectiveness ratio, SI, standard intervention; TNI, tailored navigation intervention; UC, usual care.

  2. a

    In the base case, investigator time input was 48 hours, and salaries ranged from $75,000 to $114,000 for several investigators.

Investigator time, h     
High estimate72168.61133.00915.862131.41
Low estimate24165.04111.31896.471783.81
Investigator salary, $     
High estimate86,250-131,100167.36125.42909.072009.94
Low estimate63,750-96,900166.29118.91903.261905.61
Investigator salary and time     
High combination169.41137.89920.212209.78
Low combination164.78109.68895.061757.69

Figure 1 displays the CEACs, which illustrate the probability that the SI and the TNI will be cost-effective relative to the control intervention (usual care) for a hypothetical set of decision-maker willingness-to-pay values. If the willingness to pay is $1000 per additional individual screened, then the probability that SI will be cost-effective relative to the control intervention is 0.68, and the probability that the TNI will be cost-effective relative to the control intervention is 0.11. For a $1200 willingness to pay, the probability of cost-effectiveness increases to 0.90 comparing the SI with the control intervention, and it increases to 0.56 comparing the TNI with the control intervention. In contrast, when the willingness to pay is $1200, the probability of cost-effectiveness of the TNI versus the SI is only 0.16 (results not shown). The secondary subgroup analysis comparing the control group with the TNI group that expressed a preference for screening modality and those who did not express a preference indicated a lower ICER ($906) for the group with no preference compared with the group that expressed a preference ($1473).

image

Figure 1. Cost-effectiveness acceptability curves compare usual care with the standard intervention and with the tailored navigation intervention. Blue Line: UC vs SI. Red Line: UC vs TNI. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES

Randomized trials of CRC screening navigation have focused on primary care patients who attended scheduled office visit and received a referral for screening colonoscopy. Navigators who followed “in-office” provider screening referrals achieved patient adherence rates up to 54% to 66%, which was significantly higher than controls at 13%.[30, 31] Navigation studies have also used medical records to recruit patients outside the office visit and have achieved screening rates from 27% to 41%.[32-34] The current study, in which patients were identified from an EMR, reflects comparable navigation effects.

For this study, we selected patients in CCHS practices who were noncompliant with CRC screening at baseline. From data on patients who were adherent at baseline, we estimated that the baseline screening rate in the practices was about 44%. Thus, approximately 56% of screening-eligible patients in the practices were nonadherent. In this segment of the patient population, a 33% screening rate in the SI group and a 38% screening rate among TNI participants was achieved among the patients who were nonadherent at baseline. When the SI and TNI study group screenees were added to those who were adherent with screening at baseline, the estimated overall screening rate was between 71% and 73%, which is similar to the Healthy People 2020 goals.[5]

In an earlier economic evaluation of a CRC intervention in a large, diverse patient population in Philadelphia, individuals were mailed a targeted screening invitation letter, SBT cards, an information booklet, and a reminder letter. In that study, the screening rate in the control group was 33%. The cost per additional individual screened was $319 comparing the standard mailed intervention with usual care, and the tailored intervention was less effective and more costly. Adding a reminder telephone call from a trained health educator to the tailored intervention increased the screening rate, but at a very high incremental cost. The current study, which was conducted in Delaware, provided access to screening on the basis of expressed screening test preference and encouraged screening through a tailored navigation call keyed to the individual's preferred screening test.[11] Shankaran and colleagues[35] provided a mailed brochure that was customized to the individual's name and sex to patients who had received a colonoscopy screening referral from their physician. The mailing included a reminder to schedule a screening colonoscopy, general CRC screening information, material about the importance of screening, and instructions on how to prepare for a colonoscopy. The control group had a screening rate of 59%, and those in the intervention group had a 12% increase in screening. The ICER was $43 per additional individual screened, and the intervention cost only $5 per patient. Chirikos et al[9] reported results from a study of primary care clinics serving disadvantaged populations ages 50 to 75 years in which the baseline screening rate was 12%. The primary intervention included a clinician reminder of cancer screening tests due and the division of office responsibilities to ensure follow through on tests ordered. The cost per additional individual screened was $11.53 from the societal perspective. Wolf et al[10] reported on a clinic-based Veterans Administration screening promotion program for individuals aged ≥50 years who were not compliant with screening guidelines and attended a physician visit. The baseline screening rate was 32%. The intervention included a provider workshop on CRC screening plus feedback on individual and group-specific rates. The cost per additional individual screened was $978 from a payer perspective. These cost results are similar to our findings, which included costs for planning, training staff, recruiting participants, and implementation.

We excluded the cost of participant recruitment for the control group, because this research cost is not relevant to decisions regarding the implementation of similar, future CRC screening-promotion programs. However, program managers need to recruit the target group to implement the interventions. The recruitment costs would be the same for each intervention and, thus, would not affect their relative cost-effectiveness. Recruitment costs represent from 30% up to 40% of the total cost, even in highly organized health care settings.[36, 37] Nevertheless, our average cost estimates may be an upper bound because of the higher costs associated with a research project, the small scale of the target groups, and the lack of a sophisticated EMR system to identify patient screening status.

Elsewhere, we have reported that exposure to a mailed intervention was associated with race-related disparity in CRC screening adherence that favored whites.[22] The inclusion of a relatively small proportion of nonwhite participants limits the generalizability of findings from our current study. Nonetheless, secondary analyses performed in the current study indicated that the TNI effect was substantially greater among nonwhites (SI rate, 28%; TNI rate, 42%) compared with whites (SI rate, 39%; TNI, 43%), attenuating the screening disparity. This finding suggests that the impact and cost-effectiveness of interventions intended to encourage screening may vary according to race. Future cost-effectiveness studies that consider race may be useful. Exogenous factors may have influenced the results. The state of Delaware initiated an aggressive CRC screening promotion program, and participating primary care practices engaged in various CRC screening initiatives during the study period. We assume that these initiatives had little impact on study group differences, but they may have had a modest effect on overall screening rates. Cost information is subject to the limitations of staff time input recall, overhead cost rate assumptions, inflation from delivery of the interventions in the context of a research project, and the higher costs associated with learning a new program. It would be preferable to collect time data prospectively to avoid recall errors. We caution reliance on the absolute cost estimates, although we have confidence in the relative costs of the interventions.

Adjusting investigator salaries to the levels paid to the administrators and primary care physicians who are likely to be implementing these interventions in the “real world” yields lower cost estimates for the interventions: the SI costs would decrease to $165 per participant, and the TNI costs would decrease to $275. Real-world applications should reduce costs further by using existing experienced staff, EMRs, and a larger scale of operations. It is likely that a medical assistant and a registered nurse could handle management, which would decrease average costs by $2.07 for SI and $14.68 for TNI.

In a secondary analysis (data not shown), we examined the TNI subgroups according to their preference for particular screening tests. Participants who had equal preference for SBT and colonoscopy screening were provided information and navigation for both tests, whereas those who had a screening test preference were provided information and navigation only for the preferred test. Average costs were about the same in these subgroups, but the screening rate was 0.5 in the group with equal preference compared with 0.375 in the group with a preference. The higher screening rate in the TNI subgroup with equal preference may be because they were provided information about and access to both screening modalities. This reduced the ICER to a level ($912) just above the ICER for the SI program, potentially increasing the economic attractiveness of the more effective TNI program, minus the tailoring.

Considering the cost-effectiveness acceptability along with outcomes from a randomized trial allows for a more complete evaluation of intervention effects. The economic evaluation estimates the costs and effects of the alternative interventions and the uncertainty around those estimates. Decision-makers need to consider the uncertainties, along with their willingness to pay, cost structure, and level of planning and training required to implement the interventions, to determine whether the extra resources required for navigation would be worth the investment. Decision-makers may explore ways of decreasing the costs of identifying eligible individuals and delivering tailored navigation contacts, making tailored navigation a more feasible option. In either case, the interventions are effective in increasing the screening rates above the level attained with usual care.

FUNDING SUPPORT

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES

This study was supported by National Institutes of Health/National Cancer Institute grant CA116576 and by a small grant from Olympus America. Stool blood tests were donated by Quest Diagnostics.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. FUNDING SUPPORT
  8. CONFLICT OF INTEREST DISCLOSURES
  9. REFERENCES