Patient and public involvement

Patient and community involvement is central to the design, execution and dissemination of the Hearing Norton Sound study and is described in depth in the companion protocol.34

Study population, setting and eligibility

The BSSD, comprised of 15 communities in the Norton Sound region, has been selected for the study. The study will be conducted in the school and clinic associated with each of these 15 communities. All school-aged children, from kindergarten through 12th grade, will be eligible to participate. Compared with the general US population, prevalence of hearing loss for Alaska Native children is high across this entire age range, so inclusion of all grades is important.

Recruitment

Parents of all children enrolled in BSSD will receive information about the study and a written informed consent during the school intake process. This is standard practice for health interventions provided in the school setting in Alaska. A sociodemographic survey tailored to reflect the rural Alaska environment will also be included for parental completion. Study enrolment will occur over two academic years and will happen in each school on hearing screening day. All children with signed parental consent who are present on the day of screening will be enrolled. Verbal assent will be obtained from each child in a one-on-one environment prior to participation. Verbal and non-verbal cues will be used to determine a child’s level of willingness to participate (eg, eyebrow raise indicates yes in this population). Despite parental consent, should a child indicate that they do not want to participate, they will not be enrolled. Children will be allowed to stop study involvement at any time.

We anticipate reaching 85% of target enrolment (n=1500) within the first 12 months of the trial. The intention of the study is to maintain a real-world environment where all children from kindergarten to 12th grade undergo school screening. Therefore, enrolment over two academic years is necessary to be inclusive of transferring students and new kindergartners matriculating in year 2 of the trial.

Randomisation and masking

All enrolled school-aged children in each participating community will undergo the current school screening protocol, the new mHealth screening protocol and audiometric evaluation on hearing screening day. A cluster-randomised design will be used to randomise communities to receive primary care referral or expedited telemedicine referral, such that all children from a single community will undergo the same referral process. Randomised referral assignments will be computer generated by the study statistician, stratified by four strata based on the combination of school sizes and geographic location. All possible permutations of assignment results under a stratified randomisation with block size of 2 will be generated. Selection probabilities for these possible assignments will be inversely weighted so that assignments creating greater differences of school sizes between arms (within stratum and overall) are downweighted and less likely to be generated during randomisation. The set of randomised assignments will then be randomly generated according to this weighted probability distribution.

Communities will each be assigned an identification (ID) code within their geographic strata by the leadership team in Alaska. The study statistician at Johns Hopkins will generate randomised assignments based on the community ID without the knowledge of community identities. At the day of randomisation, after all communities become eligible for the trial (ie, 50% of parental inform consents per community have been obtained), the secure-stored ID list and the independently secure-stored randomisation list will be unsealed and matched according to community ID codes to carry out the randomisation assignment.

The randomisation assignments will be kept confidential within the study team until hearing screening day, when masking the referral process assignment will no longer be possible. All study team members will be trained and reminded of the importance of randomisation concealment so that families’ willingness to participate in the trial will not be influenced by the knowledge of the referral assignment before enrolling in the study. In addition, all outcome assessors, including audiologists and ear, nose and throat (ENT) surgeons reading telemedicine consults within the Alaska Native healthcare system and study team members performing medical record abstraction, will be masked to intervention allocation. Study team members who read telemedicine consults as a part of their clinical responsibilities will abstain from reading any study-related consults. The results of the school screen, mHealth screen and audiometric assessments will also be masked, such that study team members performing the mHealth screen or audiometric assessment will be masked to the other results.

Hearing screening protocols and referral intervention

Our stakeholders and community members feel strongly that all enrolled children should derive benefit from enhanced school screening in the study and that all children in a community should undergo the same referral process. We have therefore carefully crafted a study design to honour these wishes. All enrolled children will undergo the current school screening protocol and the mHealth screening protocol, as well as a benchmark audiometric assessment, on school screening day. The referral process will be cluster randomised by community, such that all enrolled children in a community will either receive the current primary care referral process or the expedited telemedicine referral intervention. Here we describe the screening and audiometric protocols, followed by the randomised referral pathways.

Hearing screening protocols

Audiometric assessment

During screening day, each participating child will also undergo an audiometric assessment to determine the sensitivity and specificity of the current school screening and mHealth screening protocols. This assessment will be performed using Shoebox, a validated mHealth tablet audiometer (Shoebox Audiometry, Clearwater Clinical, Canada), Otometrics tympanometer (Denmark) and Otometrics Otocam digital otoscope (Denmark). Study team members who are trained audiologists will perform this assessment. They will be blinded to the results of the mHealth and school screens. Otoscopy is important for diagnosing ear conditions requiring treatment in a population with a high prevalence of otitis media, but it requires clinical skills and training that are not practical for implementation in a large-scale screening programme. We have therefore included otoscopy within our audiometric assessment as a component of the benchmark evaluation instead of incorporating it into the mHealth screening itself.

• Air and bone conduction pure-tone audiometry using a validated mHealth tablet audiometer at 0.5 kHz, 1 kHz, 2 kHz and 4 kHz.43 A referral will be generated for PTA >25 dB or a threshold >30 dB in a single frequency. Bone conduction will be performed at 0.5 kHz, 1 kHz, 2 kHz and 4 kHz if hearing loss reaches the threshold for referral. Prevalence of hearing loss (PTA >25 dB at 0.5 kHz, 1 kHz, 2 kHz and 4 kHz) will be calculated from this audiogram.

• Diagnostic tympanometry. Type B (flat) tympanogram or negative pressure <−200 daPa will generate a referral.

• Digital otoscopy. Digital images of bilateral tympanic membranes will be taken. Pathological findings will generate a referral at the discretion of the audiologist.

Schools will generate a list of all children referred by the current screening protocol, as is already standard practice in each school. The study team will generate a list from the mHealth screen of all enrolled children who refer either by pure-tone screening or tympanometry. If a child passes the current and mHealth screening protocols but is noted during the audiometric assessment to have conditions requiring treatment, they will also be flagged for referral. Referral lists will be reconciled at the conclusion of the screening day to ensure that all families of children who refer will receive a single notification. Once a reconciled referral list has been developed by the study team, it will be handed over to local school leadership and clinic staff to initiate and manage the referral process. This hand-off is done to maintain a real-world environment that enhances generalisability of the study for future implementation in Norton Sound and other regions of Alaska.

Intervention arms

Communities will be randomised to the current primary care referral process or expedited telemedicine referral. Figure 2 highlights the complexity of the diagnosis and treatment pathway for childhood hearing loss in rural Alaska communities, with multiple branch points, wait times and potential need to travel for higher levels of care. The key differences between current (figure 2A) and expedited telemedicine referral processes (figure 2B) can be summarised by two fundamental issues we have identified with the current process: (1) children who refer screening can be lost to follow-up, preventing them from ever entering the healthcare system for diagnosis and treatment, and (2) despite the integration of telemedicine in village health clinics for many years, telemedicine has never been systematically applied to expedite school hearing screening referral.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Primary care (A) and expedited telemedicine (B) referral pathways for diagnosis and treatment of childhood hearing loss in rural Alaska. CHA, community health aide; ENT, Ear, nose and throat surgery. 

Current primary care referral process

In communities randomised to the current primary care process, families will be notified if their children screen positive in exactly the same method each school had been using previously. This process is managed by the school and involves a letter home to the parents, either sent with the child or by mail, requesting that the parent/caregiver bring the child to the village health clinic for evaluation by a CHA. This assessment can lead to multiple treatment pathways, including waiting for audiology field clinic, telemedicine consultation to audiology or referral to a primary care physician. ENT surgery becomes involved in the treatment pathway if recommended by audiology (figure 2A). In addition to letters home requesting that parents bring children into clinic for evaluation, a list of referred children is also given to the Department of Audiology at Norton Sound Health Corporation. Audiology staff reach out to families to schedule appointments during the next available audiology clinic.

Expedited telemedicine referral

In communities randomised to the expedited telemedicine intervention, parents of children who screen positive will receive a phone call from the school or the clinic notifying them of the day and time of their child’s telemedicine consultation appointment. Appointments will be made with CHAs who have dedicated time blocked off to perform referral telemedicine consults to audiology (figure 2B). Participating children who refer will be transported to clinic for their appointment with adult chaperones. Parents will be encouraged but not required to attend, except for children grades 2 and younger, for whom parental participation will be required. Thus, the expedited telemedicine referral pathway will systematically apply telemedicine consultation to audiology for all referrals and does not require the presence of a parent for completion. Similar to the current primary care referral process, ENT surgery will become involved in the treatment pathway if recommended by audiology.

Schools will manage the expedited telemedicine referral process jointly with clinic staff. The role of the study team is simply to provide a list to local school leadership and clinic staff of students being referred.

Non-participating children in communities assigned to the expedited telemedicine intervention arm will receive standard referral following the current primary care referral process. All services provided at the clinic for the purposes of clinical care for Alaska Natives, including the telemedicine intervention, will be covered by Indian Health Service.

It is important to note that all communities participating in this study have telemedicine within village health clinics that is routinely used by CHAs for triage and management of health problems, including ear-related complaints. While telemedicine is not routinely used for school hearing screening referrals at the current time, it will remain available to CHAs to use at their discretion throughout the course of the study, maintaining a real-world environment where some crossover between the study arms may occur. We will monitor the use of telemedicine through chart reviews in the multiorganisational electronic medical record (EMR) used by Norton Sound Health Corporation that will be conducted on all participating children who refer. Chart reviews will be completed within 9 months of the date of school hearing screening. If we find a child has been lost to follow-up through chart review, we will contact the families at that time and expedite diagnosis and treatment as needed.

Outcome measures

Secondary outcomes

Sensitivity and specificity of the screening protocols will be measured against a benchmark audiometric assessment, including air and bone conduction audiogram, tympanometry and digital otoscopy.44–47 The accepted definition of hearing loss from the WHO will be used (PTA >25 dB over 0.5 kHz, 1 kHz, 2 kHz and 4 kHz).48 We have chosen to use the standard WHO definition of hearing loss instead of disabling childhood hearing loss (PTA >30 dB over 0.5 kHz, 1 kHz, 2 kHz and 4 kHz) because of ample evidence that mild hearing loss has a detrimental impact on language outcomes, school performance and vocational opportunities.1–9 We hypothesise that the mHealth screening protocol will be more sensitive than the current school screening protocol, more accurately identifying children with hearing loss because it includes a middle ear assessment that is essential to fully evaluating ear health in a population with high prevalence of otitis media (hypothesis 2).

Prevalence of hearing loss as defined by the WHO (PTA >25 dB at 0.5 kHz, 1 kHz, 2 kHz and 4 kHz in either ear) will be compared from the first and second annual audiometric assessments.48 We hypothesise that hearing loss prevalence will decrease in telemedicine referral communities compared with primary care referral communities due to expedited diagnosis and treatment (hypothesis 3a).

Hearing-related quality of life will be measured using HEAR-QL child self-report.49 50 HEAR-QL has been selected, because it is the only validated tool available for hearing-specific quality of life in children 7–18 years of age. This study represents the first use of HEAR-QL in Alaska Natives and an opportunity to extend utilisation of this instrument to the Alaska Native population. We have obtained feedback from our initial focus groups on consequences of hearing loss that are specific to Alaskan Native culture. Based on this feedback, we have developed a short supplemental section to the HEAR-QL questionnaire to capture culture-specific quality of life (eg, willingness to participate in activities such as hunting or use of motor vehicles). We hypothesise that hearing-related quality of life measures will improve from year 1 to year 2 in telemedicine referral communities compared with primary care referral (hypothesis 3b).

School performance will be assessed using AIMSweb, a validated, reliable measure of math and reading performance that is administered to all children in kindergarten to 8th grade three times annually in BSSD.51 52 We hypothesise that school performance will improve from year 1 to year 2 in telemedicine referral communities compared with primary care referral communities due to expedited diagnosis and treatment (hypothesis 3c).

Timing of data collection

The timing of data collection is detailed in figure 3. The qualitative components of the study illustrated in figure 3 are discussed in the companion protocol.34 The study team will administer the mHealth screening, audiometric assessment and HEAR-QL in every community once annually for two consecutive academic years. AIMSweb is administered by schools three times annually. These data will be released to the study team as it becomes available.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Timing of qualitative and quantitative data collection in the Hearing Norton Sound comparative effectiveness community randomised trial. EMR, electronic medical record; ICD-10, International Statistical Classification of Diseases, 10th Revision; mHealth, mobile health.

Sample size

There are approximately 1900 students enrolled in the BSSD across 15 communities (BSSD, personal communication, 2016). We anticipate that total accrual in the study of 80% of children who attend BSSD schools (n=1520) is feasible because the school district already participates in state-mandated annual school hearing screening. Once 50% of parental consents have been obtained in all communities, the randomisation process will occur so the research team can plan for logistics for the referrals. This study is intended to be inclusive of all children in BSSD, and thus we will continue to accept parental consents until school screening day, which represents the last day the randomised referral assignment can be strictly concealed. All children who assent to participation, have parental consent and are present on the day of screening will be enrolled. Maximum enrolment is therefore expected to be 1900 children.

Power calculations were performed for an assumed fixed sample size (n=1500). For the primary outcome of time to diagnosis, our preliminary studies have indicated that wait times decreased from an average of 4.2 months (18 weeks) to 2.5 months (10 weeks) when telemedicine was implemented in village health clinics in the Norton Sound region, with a SD of 0.7 months.16 Based on these data, we hypothesise that expedited telemedicine referral will reduce wait times by 8 weeks compared with current primary care referral. Assuming 80% power, type 1 error of 0.05, intracluster correlation coefficient (ICC) of 0.25 and 10% loss to follow-up, 15 clusters will provide 81% power to assess a difference in absolute mean reduction in time to diagnosis of 3 weeks between arms. If we assume ICC=0.05, then we will have 80% power to detect a difference in mean reduction in time to diagnosis of 1.5 weeks between arms. In general, we will have sufficient power to detect a mean difference in effect size of 0.8 and 0.4 between arms for any continuous secondary outcomes assuming ICC=0.25 and 0.05, respectively. Assuming 20% prevalence of hearing loss (300 children with confirmed hearing loss by audiometric assessment) and type I error=0.05, we will have 80% power to detect a sensitivity of 96.6% in the mHealth screening versus 90% in the current screening protocol, with assumed ICC=0.05.

Censoring and missing data

As we plan to screen all children in all grades annually with the current school screening protocol and mHealth screen, in year 2, the 12th graders from year 1 will no longer be enrolled in school, and a new kindergarten class will have matriculated. This natural progression in each academic year is representative of the school-aged population at large. Due to this design consideration, these children (~15% of the study population) will contribute only up to 1 annual data point, with the other annual data point designated as a protocol-defined censoring event that will be considered missing at random so will not induce bias with proper analysis. Children in other grade-years (~85% of the study population) will contribute up to 2 annual data points.

For the primary outcome of time to diagnosis, we will capture and review all encounters documented in the multiorganisational EMR used by NSHC to determine whether a participating child sought care after screening positive (a binary outcome), and among those who sought care, the time to first ear/hearing ICD-10 diagnosis. We believe we can achieve complete data collection without any missing data for these analyses and will use mixed effects modelling to address the within-child outcome correlation for children who contributed 2 annual data points in the same part of the analyses.

For sensitivity, specificity and prevalence estimates, the data on which we base the analyses will be available for all children with parental consent who were present at school on hearing screening day. We will carefully document and compare characteristics of school absence and dropout of those who consented and participated to those who were absent for the screening and will conduct sensitivity analyses on the prevalence estimates using multiple imputation and propensity score weighting to evaluate the robustness of results. We will also monitor absenteeism in real time during each of the four enrolment periods (fall 2017, spring 2018, fall 2018 and spring 2019). If at the end of an enrolment period missing data due to absenteeism exceeds 10%, a make-up screening will be performed in communities where it is needed.

While school hearing screening is mandated by the state of Alaska, there is currently no policy in place regarding missed screening due to absenteeism from school. In some schools, rescreening is done, and in others, it is not. We plan to rescreen communities with >10% absenteeism, with the goal of creating a rescreening policy that could be incorporated into the Alaska school screening mandate if the intervention is successful and ultimately becomes standard of care in the state.

Statistical significance and software

Two-sided p values <0.05 will be considered statistically significant. All quantitative analyses will be performed with Stata V.13.1 and SAS V.9.4.

Intention to treat

All participating communities have telemedicine capabilities within village health clinics that will remain available to CHAs to use at their discretion throughout the course of the study, maintaining a real-world environment where some crossover between the study arms may occur. Data will nevertheless be evaluated using an intention-to-treat approach, with all data from intervention and control communities assessed by allocation regardless of adherence to the intervention arm. We will monitor the use of telemedicine through EMR chart reviews that will be conducted on all children screening positive.

Overview of analysis

Individual child and community characteristics will be compared to identify potential imbalances of key variables between randomised arms, and any imbalances in characteristics noted during the initial analysis will be used as adjustment variables in subsequent main analyses and other multivariable modelling. Sociodemographic information collected by parental self-report during the school intake process will facilitate any necessary confounding adjustment. Reliability of self-reported data will be confirmed in a random subset of questionnaires by comparison with socioeconomic data kept on file at Norton Sound Health Corporation. This test for reliability was selected because other individual-level surrogate measures of SES, such as qualification for free or reduced-price lunch or Medicaid eligibility, are nearly universal in participating communities and are therefore not useful for differentiating SES in the study population.

Primary analyses

The primary outcome, time to ICD-10 ear/hearing diagnosis from the date of screening, will be calculated in days from screening day. If there exist children who were lost to follow-up after screening positive, data analyses will proceed with a two-part approach: first, the per cent not receiving care will be compared between the two referral arms; next, conditioning on those who sought care, the mean time to diagnosis will be compared between the two referral groups. Outcome correlation for children who contributed 2 annual data points will be addressed using random-effects frailty models for correlated time-to-event outcomes, mixed effects modelling for continuous outcomes, and generalised estimating equations (GEEs) for binary outcomes to assess the marginal effects of the intervention. Annual time trend in mean outcome between arms will be captured by modelling the main effects of intervention and time and the interaction effect of intervention by time. Children contributing 1 or both annual data points will be included in this unified modelling approach for data analyses.

Secondary analyses

Sensitivity and specificity of the current and mHealth screening protocols will be calculated by comparing to the benchmark audiometric assessment.

Hearing loss prevalence will be calculated by community separately for year 1 and year 2 of screening. Effect of the intervention on changes in hearing loss prevalence between the two time-points will be assessed using GEE models incorporating a time by intervention interaction term, other appropriate covariates and a correlation structure reflecting both community clustering and temporal correlations using all available data. Robust estimation will be used for inferences on the marginal effects.

HEAR-QL will be compared between children with hearing loss and their normal-hearing peers. Effect of the intervention on trajectory of HEAR-QL scores over time between arms will be assessed using mixed effects models that properly account for temporal correlation and clustering by communities. To the extent possible, we will correlate HEAR-QL outcome with ear and hearing diagnosis, AIMSweb math and reading scores and other relevant outcomes to evaluate validity of this tool for Alaska Native populations.

AIMSweb math and reading scores will be compared between children with hearing loss and their normal-hearing peers. Effect of the intervention on trajectory of AIMSweb scores over time between arms will be assessed using mixed effects models that properly account for temporal correlation and clustering by communities.

Heterogeneity of treatment effects

In this rural Alaska population that experiences such high prevalence of middle ear disease, it is possible that previous hearing loss diagnosis and treatment may influence efficacy of the intervention. We have defined an ear/hearing condition under active management as an audiology or ENT encounter within 3 months of the date of screening or a child wearing a hearing aid on screening day. Previously managed conditions are defined as an audiology or ENT encounter >3 months prior to screening day. We will evaluate for possible heterogeneity of treatment effect (HTE) by analysing outcomes separately for children with new/previously undiagnosed conditions, actively managed conditions and previously managed conditions. The qualitative interviews described in the companion protocol will be essential for fully understanding the impact of prior diagnoses on families’ response to follow-up for hearing screening referrals.

Aside from previous hearing loss diagnosis, the other main subgroup analysis for exploring HTE will be native versus non-native subgroups. We will also explore for potential HTEs by SES variables such as age, gender, caretaker education and environment. We have no other planned subgroup analyses, as the intervention is intended to be applied at the community level and not to individual subpopulations.

If any potential HTE was found, a formal statistical comparison testing treatment group by subgroup interaction through statistical modelling will be carried out to confirm statistical significance of such HTE. Subgroup treatment effects and corresponding 95% CI will be estimated. We will also carefully examine subgroup characteristics for differences that may be related to such HTE as part of a hypothesis-generating exercise. We will interpret any potential HTE finding cautiously, including reporting the number of post hoc analyses conducted to arrive at such a finding.