This chapter intends to familiarize the reader with the steps involved in conducting research in the area of noise and music and their impact on the health and well-being of human beings. Research primarily intends to create new knowledge to solve challenging healthcare situations. With this perspective, research and knowledge creation will be used as synonyms. The seven main steps in knowledge creation are depicted in the schematic diagram in Figure 22.1. The first step is identifying a challenging healthcare situation in the domain of deleterious health effects of noise and beneficial effects of music. The challenge is stated as a research question in nine primary domains (description in small groups, evaluating laboratory-based parameters, estimating prevalence/incidence in populations, evaluating therapies, measuring costs, developing tools to screen/predict/diagnose, measuring risk factors, predicting variables by estimating the strength of correlation and examining beliefs/perceptions/experiences). From the research question are derived the objectives. The second step is to develop a conceptual framework of variables around the objective. The third step is to review existing literature to examine if any researcher has already addressed the objective. The fourth step is to choose the most appropriate study design. The fifth step involves selecting the sampling strategy and calculating the sample size followed by data collection. The sixth step is summarizing data to arrive at results. The last and seventh step is to apply the tests of significance to examine the generalizability of the findings to the target population.1 Internationally accepted standards for reporting research studies employing various study designs will be elaborated.
• Understand the process of identifying a challenging healthcare situation in the domain of noise, health and music, and state it as a research objective.
• Construct a conceptual framework mapping all the variables.
• Review literature to search if an answer exists for the objective based on previous research.
• Design a study to answer the research objective.
• Choose a representative sample from the universe of the study.
• Collect and summarize the data.
• Perform appropriate tests of significance to generalize the findings for the universe of the study.
Clinical immersion to identify challenging healthcare situations requiring knowledge creation is the first step. Ethnographic observation over a period of three weeks should be conducted in a structured manner. To gain broad perspective stakeholders from allied branches should also be involved. The team of observers depends on the context. For music-related situations the team can consist of musicians, sound engineers, psychologists, ear nose throat specialists and audiologists. For identifying challenges in organized and unorganized sectors of industries, the team may consist of human resource managers, workplace quality enforcers, health activists, lawyers, psychologists, ear nose throat specialists and audiologists. In healthcare settings medical personnel of the concerned specialty or hospital area, human resource managers, quality control department staff, psychologists, ear nose throat specialists and audiologists can be involved. Identifying challenges in community settings requires local opinion leaders, gram panchayat members (rural areas), ward councillors (urban areas), schoolteachers, professional social workers, psychologists, ear nose throat specialists and audiologists. After identifying the challenges, the team should have a focus group discussion with the people exposed to high noise levels in these settings. The list of challenges should be displayed; the exposed people should be requested to examine their relevance and indicate areas in which they feel research should be undertaken. The high-priority challenges identified by this technique should be chosen to create new knowledge. This is called researching with the people.
The next step after identifying the critical challenges is to frame a research question and identify the category to which it belongs. Readers may download a free education app named Research Genie from Google Play Store or App Store that describes questions in all the nine domains relevant to healthcare. A three-series article describes the app and method to utilize it to frame research questions in these domains. Other apps that may assist with research are Academia.edu, R Discovery, PubMed, PMC books, Research Notes (contains a large open access library for review of literature), ResearchGate (assists in connecting to peers with similar research interests), Endnote and Paperpile (assists in organizing research documents). The research question should consist of some or all of the following components: population that is being studied (P), interventions if they are being evaluated (I), comparator group if comparisons are being made (C), outcomes (O) and time period by which the study has to be completed.2
The first category of questions is description in small groups. It deals with the proportion of a particular event. For instance, what is the proportion of noise-induced hearing loss (outcome/event) in people working in a pub located in a particular area of city (population)? The answer to this question will give us estimates of the burden of noise-induced hearing loss in a particular pub identified by the researcher. Based on this, large-scale population-based studies on incidence (new cases) and prevalence (existing cases) can be conducted to estimate population-level disease burdens or event rates.
The second category of questions is about laboratory-based values. Its structure is similar to the first category. The difference is that the outcome here is a number. What is the range of noise levels (outcome/event) generated during granite polishing in a particular construction site located in Jaipur (population)? The answer to these questions assists in getting estimates of a numerical measure in a particular context. If the values are found to be clinically relevant, normative ranges can be evaluated by conducting studies in the concerned population.
Estimating the prevalence/incidence in a community is the third category. What is the prevalence of noise-induced hearing loss (outcome/event) in individuals in the urban wards of a district located near an airport (population)? This question deals with prevalence. What is the incidence of noise-induced hearing loss (outcome/event) in newly recruited health personnel working in the laundry section of hospitals (population) over a period of three years? Here the incidence of new cases is measured. The answer to this category of questions assists in policy and advocacy to take actions to prevent noise-induced hearing loss.
The fourth type of question compares two methods of interventions. In people exposed to high industrial noise (population), what is the comfort level (outcome/event) using an insert type of ear protection device (intervention 1) versus head worn ear protection device (intervention 2)? The answer to these questions aids in choosing the most appropriate intervention.
Measuring costs is the fifth category of questions. What is the cost effectiveness (outcome/event) of pre-employment screening for hearing impairment (intervention) in industries with high noise to prevent compensation claims? The estimation of costs assists in prioritizing resources for a particular activity over another.
Research questions based on developing tools to screen/predict/diagnose events is the sixth category. Can we use a smartphone-based decision-making application (new test/device/paradigm/position) to predict the probability of mobile phone users (population) developing noise-induced hearing loss (outcome/event) is an example of a question of this category. These questions are directed at creating innovations in the area of noise and health.
The seventh category of questions deals with measuring risk factors. Are smokers (at-risk group) more likely to have develop noise-induced hearing loss (outcome/event) compared to those who do not smoke (not-at-risk group)? Identifying and quantifying risk guides us to develop risk-reduction strategies.
Can we predict one number based on the other is the eighth type of question which is based on mathematical correlation. Is noise dosage (quantitative parameter 1) correlated to decibel dip on pure tone audiogram (quantitative parameter 2)? The answers to these questions will help arrive at the formula to predict one based on the other.
The last and ninth type of question deals with describing beliefs/perceptions/experiences. Why do people working in high-noise areas not wear ear- protection devices is a question that intends to explore beliefs, perceptions and experiences about wearing ear-protection devices. The answer to these questions uncovers the basis of health-related behaviour and thereby gives access for changing behaviour.
After framing the research question, a conceptual framework of all the known factors that influence the outcome should be mapped and defined. This will be explained using an example. Consider the research question that compares two types of interventions. In people exposed to high industrial noise (population) what is the comfort level (outcome/event) using an insert type of ear protection device (intervention 1) versus head worn ear protection device (intervention 2)?
Here the outcome is comfort level which is called the dependent variable as it is dependent on the interventions, namely, insert and head worn ear protection devices. The researcher should clearly define the dependent and independent variables. In certain situations where two groups are not compared, there will be only one variable which is the outcome. All the other variables influencing the results must be mapped around the dependent and independent variables. These variables are termed confounding factors, or confounders, and effect modifiers. Confounders influence both the dependent and independent variables, whereas effect modifiers influence only the dependent variable. For example, the age of the person will influence the comfort levels as well as utilization of the ear protection device. Pre-existing hearing loss will influence comfort levels due to a phenomenon called recruitment (sudden disproportionate increase in perception of loudness with small increments in noise levels). So age is a confounder and pre-existing hearing loss is an effect modifier. Both these variables can distort the results of the study. After mapping all the variables, each variable should be classified as numeric (numbers), nominal (categories) and ordinal (grades). The type of ear protection device is a nominal variable; comfort levels measured using a psychometrically validated scale is a numerical variable; and classifying the degrees of comfort as “no discomfort –- mild discomfort – moderate discomfort – severe discomfort – painful to wear” is an ordinal variable. After operationalizing the variables we can proceed to the next stage, that is, stating the hypothesis and defining the objectives.
Hypothesis is stating the research question for statistical testing. Statistical concepts will be elaborated in the section on “applying statistical tests of significance.” An example will be used to illustrate templates that frame objectives from the research question. The readers may use these templates to define their objectives. Table 22.1 shows research questions in each healthcare domain related to noise and the objectives derived from these questions.
|Research domain||Research question||Objective|
|Description||What is the proportion of noise-induced hearing loss (outcome/event) in people working in a pub located in a particular area of city (population)?||To estimate the proportion of noise-induced hearing loss (outcome/event) in people working in a pub located in a particular area of city (population)|
|Lab Range||What is the range of noise levels (outcome/event) generated during granite polishing in a particular construction site located in Jaipur (population)?||To estimate the range of noise levels (outcome/event) generated during granite polishing in a particular construction site located in Jaipur (population)|
|What is the prevalence of noise-induced hearing loss (outcome/event) in individuals in urban wards of a district located near an airport (population)?
What is the incidence of noise-induced hearing loss (outcome/event) in newly recruited health personnel working in the laundry section of hospitals (population) over a period of three years?
|To estimate the prevalence of noise-induced hearing loss (outcome/event) in individuals in urban wards of a district located near an airport (population)
To estimate the incidence of noise-induced hearing loss (outcome/event) in newly recruited health personnel working in the laundry section of hospitals (population) over a period of three years
|Therapy||In people exposed to high industrial noise (population) what is the comfort level (outcome/event) using insert type of ear protection device (intervention 1) versus head-worn ear protection device (intervention 2)?||To compare comfort level (outcome/event) in people exposed to high industrial noise (population) using insert type of ear protection device (intervention 1) versus head-worn ear protection device (intervention 2)|
|Cost||What is the cost effectiveness (outcome/event) of pre-employment screening for hearing impairment (intervention) in industries with high noise to prevent compensation claims?||To compare the cost effectiveness (outcome/event) of pre-employment screening for hearing impairment (intervention) in industries with high noise to prevent compensation claims|
|New Test||Can we use a smartphone-based decision-making application (new test/device/paradigm/position) to predict the probability of developing noise-induced hearing loss (outcome/event) among mobile phone users (population)?||To compare the accuracy of decision-making using a smartphone-based application (new test/device/ /paradigm/position) to predict the probability of developing noise-induced hearing loss (outcome/event) among mobile phone users (population)|
|Risk Measurement||Are smokers (at-risk group) more likely to have develop noise- induced hearing loss (outcome/event) in comparison to those who do not smoke (not-at-risk group)?||To estimate the risk of developing noise-induced hearing loss (outcome/event) in smokers (at-risk group) in comparison to those who do not smoke (not-at-risk group)|
|Correlation||Is noise dosage (quantitative parameter 1) correlated to decibel dip on pure tone audiogram (quantitative parameter 2)?||To estimate the strength of correlation (outcome/event) between noise dosage (quantitative parameter 1) and decibel dip on pure tone audiogram (quantitative parameter 2)|
|Why do people working in high-noise areas not wear ear protection devices is a question that intends to explore beliefs, perceptions and experiences about wearing ear protection devices||To describe the perceptions of people working in high-noise areas regarding use of ear protection devices|
A comprehensive review of literature is a critical step that must be completed before embarking on designing the study to answer the objective. A strategy should be employed to review existing knowledge base. The most commonly used database is PubMed. The database should be searched using medical subject headings (MeSH). Various methods of advanced search strategies are available to extract relevant citations from this database. Other commonly used databases are Cochrane Library, Google Scholar, Indian Medlars, Embase, NHS Evidence and CINAHL. A reference manager software like Refman, Endnote or Zotero should be utilized to store and cite the references. A researcher working in the domain of noise and health should constantly update the references on a quarterly basis to understand the latest developments.
If literature review reveals that no prior work has been done to answer the research question, then the study can be designed to create new knowledge to address the issue. The study design is chosen on the basis of the type of research question and there are guidelines for each design.
Descriptive, lab range and incidence/prevalence questions: The study design best suited to answer these questions is a cross-sectional study (observational design). It can be prospective or retrospective. The quality of these types of studies is evaluated using the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines available in the website of Equator Network.3
Therapy questions: This category of question is best answered by randomized controlled trials (RCT). The main intention of randomization is to eliminate selection bias and achieve balance of prognosis. Human beings by design are biased either intentionally or unintentionally. The Consolidated Standards of Reporting Trials (CONSORT) is the most appropriate guideline to ensure quality of randomized controlled trials.4
Cost questions: Measuring costs in healthcare is elaborated by the term ‘health economics’. There are primarily three types of economic analysis, namely, cost effectiveness analysis, cost utility analysis and cost benefit analysis. Cost effectiveness analysis (CEA), measures cost incurred per unit change of a clinical parameter (e.g., rupees spent to reduce one decibel of noise). The results of the CEA guide the policy maker to choose the most effective intervention. Cost utility analysis estimates the cost to change health utility measures in a given context. Disability adjusted life years (DALYs) is an example of a utility measure. It is a measure of the individual burden of disability derived from large scale population studies. These measures are available from the Global Burden of Disease study. Many countries have developed national specific utility measures. Suboptimal management of noise-induced hearing loss can lead to hearing disability in the long term. So the cost to reduce one DALY by various interventions in high noise environments is an example of cost utility analysis. As utility measures are standardized across populations, a cost utility analysis can be used to compare various interventions across different contexts. These comparisons are tabulated as Stochastic League Tables. Cost benefit analysis measures the economic benefit that an individual gains through his earning in terms of cash income for the amount spent for the health condition. The Consolidated Health Economic Evaluation Reporting Standards (CHEERS) guideline is the most comprehensive method to ensure the quality of health economic studies.5- 7
New test questions: Validations design is the most appropriate study design to answer this category of questions. Here various characteristics of the new tool are estimated based on the reference test. Different characteristics of the new test are described in terms of validity, reliability, sensitivity, specificity, positive predictive value, negative predictive value, accuracy, precision, efficiency and likelihood ratio.
Validity is a measure of whether the instrument measures what it is supposed to measure. Reliability is a measure of the consistency with which a research instrument yields results. Expressed in simple terms, a reliable instrument is one which yields the same results every time it is applied to the same subject by different testers (intertester consistency) or to different subjects by the same tester (intratester consistency). Sensitivity is a measure of how well the test detects a condition. Specificity measures how well the test excludes people without the disease. Positive predictive value is a measure of the probability of the disease being present if the test is positive. Negative predictive value is a measure of the probability of disease not being present if the test is negative. Accuracy and precision measure what proportion of tests will give the correct result. So it is a combination of sensitivity and specificity. The likelihood ratio of positive test is a measure of how much more likely the test will be positive in a person with disease compared to a person without the disease. We recommend you learn these concepts in detail if you plan to conduct validation studies. Once you understand the concepts, these values can be calculated using online calculators. The recommended guideline to ensure quality of validation studies is Standards for the Reporting of Diagnostic Accuracy studies (STARD).8; 9
Risk measurement question: In this type of study two types of designs can be used. Case-control and matched cohort designs are the most appropriate study designs to answer these questions. In case-control design the event has occurred and for each case an age and gender matched control is chosen. In matched cohort designs, two cohorts, one with exposure and another without, are followed up to observe the occurrence of the event. Case-control and cohort designs are classified as observational analytical studies, so the quality of these type of studies are evaluated using the STROBE guidelines.
Correlation question: The data for correlated parameters are collected from observational studies. So the design in these type of objectives is observational analytical. It is analytical because the two variables are correlated. Here again STROBE guidelines are most appropriate to evaluate the quality.
Beliefs/Perception/Experience questions: These type of objectives explore an entirely different domain in comparison to all the previous objectives. In these situations, in-depth interviews, focus group discussions and ethnographic observations are employed to describe beliefs, perceptions and live experiences of people in various contexts. The interviews are recorded and transcribed. Themes are derived from the written transcripts. In another approach, theories grounded in the data can be conceptualized. All these designs are called qualitative study designs. Adequate training is required for conducting these types of studies. Though the data may appear abstract, specific guidelines have been developed to evaluate the quality of these study designs. The Standard for Reporting Qualitative Research (SRQR) is a comprehensive checklist for ensuring the quality of these types of designs.10; 11
The main intention of sampling strategy and sample size calculation is to ensure that the selected sample is representative of the target population. Sampling strategy is based on inclusion and exclusion criteria. Inclusion criteria specify all the requirements to select the correct target population. Exclusion criteria specify those aspects which may confound the results hence cases with these factors should be excluded. Take the therapy type of research question: In people exposed to high industrial noise (population) what is the comfort level (outcome/event) using an insert type of ear protection device (intervention 1) versus head worn ear protection device (intervention 2)? Here the inclusion criteria would be people exposed to high industrial noise who wear ear protection devices. The exclusion criteria would be those with allergic otitis externa (allergy of the ear canal). These cases need to be excluded as the skin condition could cause discomfort which would confound the outcome. In other words, comfort levels with using the ear protection device could be influenced by the ear canal condition and give erroneous results. There are two types of sampling strategies, namely, probability sampling and non-probability sampling. In probability sampling every individual has an equal chance of getting selected. Non-probability sampling is mainly based on the convenience of the researcher.
Sample size is determined by the extent of variability in the population in the parameters being measured. If two parameters are being compared, the expected difference between these parameters influences the sample size. So in research objectives in which observational designs are employed, variability is determined by examining previous studies. When comparisons are being made, the researcher should have estimates of the expected difference based either on their experience or on a pilot study.12 The reader is requested to use the Research Genie app to understand sample size calculation for each objective in the nine healthcare domains elaborated in the previous sections.
Data is collected according to an objective. There are primarily three main families of data, namely, numerical, categorical and qualitative. Numerical data is expressed as numbers on a scale with zero as reference. Sound pressure levels measured as pascals is a numerical measure. The common measure used to depict noise levels is decibels. Decibel is a logarithmic scale. For an arithmetic function with a decibel-based measurement, the latter has to be converted to sound pressure levels. This is an important concept for researchers in the domain of noise to understand. Categorical data can be classified as distinct groups. Insert type and head worn type of ear protection devices is an example of categorical data. Qualitative data is an entirely different family of data. Structured responses to questions are mistakenly considered as qualitative data. Qualitative data is written narratives, for example, the narrations of users of ear protection devices such as “It causes a feeling of discomfort around the ear”, “Sometimes it pains” and “I do not like to wear it”. So the data collection proforma should be designed according to the type of data, with appropriate codes for categorical data. The collected data has to be cleaned for any error before analysis.
The collected data is summarized to make it meaningful. A collection of numbers or scripts cannot be used to make decisions. For example, a long list of minute-by-minute noise samples over a 24-hour period has no meaning. The average value, the most common value or the centrally placed value assists in interpreting quantitative data (numerical or categorical). For qualitative data a thematic framework assists in understanding the transcripts. For each type of objective a specific method is employed to summarize data. Research Genie can assist in identifying the most appropriate summary measure for a given objective.
The findings of the study are based on the sample selected from the universe where the research question was asked. Sampling and sample size ensure representative sampling, which makes the findings generalizable to the universe. The application of the tests of significance derives a probability measure of this generalizability. A particular objective has a specific test of significance. A simple approach to choose the test of significance is to understand it in the context of each objective in the nine domains of healthcare. For the description and incidence/prevalence type of objective, the Z-test of proportions is the most appropriate test of significance. Here we will introduce the concept of assumptions before applying the test of significance. When we communicate to a person in Kannada, for example, and expect a response, we assume that the person understands Kannada. Likewise, data has to fulfill a certain set of assumptions before the Z-test of proportions can be applied. Random sampling and independence of measurement (measurement obtained in an individual should not be dependent on any other prior factor that can systematically bias the measurement) are two assumptions to be fulfilled before applying the Z-test of proportion. The assumptions ensure that data analyzed by the computer program is representative of the population/universe. Computers cannot ensure random sampling and independence of measurement; it has to be done by the researcher. The Z-test of means or student’s t-test is the correct test of significance for the lab range based objective. To test statistical significance of two interventions or to compare costs, an independent sample t-test is employed if data is normally distributed and the Mann-Whitney U test or Wilcoxon signed-rank test if the data is skewed and assumptions violated. The tests employed when assumptions are violated are less powerful in assisting us to generalize the results to the universe. If three interventions are compared, analysis of variance (ANOVA) is used if data is normally distributed and the Kruskal Wallis test is used if assumptions are violated. For objectives that measure risk chi-square is employed if assumptions are fulfilled and Fisher’s test if assumptions are violated. Pearson’s correlation is used for normally distributed data and Spearman’s for data that violate assumptions in the correlation category of questions. Objectives based on beliefs, perceptions and experiences are answered by qualitative methods that do not employ tests of significance.
A checklist to maintain quality in research studies on noise, music and health is placed below.
• Conceptual framework: Has a comprehensive framework of all variables influencing the outcomes been constructed?
• Review of literature: Is the search strategy robust enough to scan all existing knowledge bases?
• Study design: Has the most appropriate study design been chosen?
• Sampling: Have all the aspects of heterogeneity in the universe of interest been considered while sampling to ensure representativeness?
• Data collection and summarizing: Is the data collection platform accurate and of quality? Has the most appropriate summary measure been chosen?
• Applying tests of significance: Has the correct test of significance been employed to generalize the results to the universe of the study?
This chapter creates a robust framework for undertaking knowledge creation in the domain of health, noise and music. The main conclusions it draws are as follows.
• Knowledge creation by research is required to address the challenges in the domain of noise, music and health.
• There are seven steps in the process of knowledge creation, namely, identifying a challenging healthcare situation in the domain of deleterious health effects of noise and beneficial effects of music and deriving objectives, developing a conceptual framework of variables around the objective, reviewing existing literature to examine if any researcher has already addressed the objective, choosing the most appropriate study design, selecting the sampling strategy and calculating the sample size followed by data collection, summarizing data to arrive at results and applying the tests of significance to examine the generalizability of the findings to the target population.