Reading Response Discussions

Hello all,

I hope this is what the assignment was about. I by mistake read Arcaya’s article but liked it and thought, I could still use their study for inspiration.

1. Reading the Pagidipati article, I wasn’t surprised to see that in some countries with a high HIV stigma, family members try to not have AIDS as the cause of death. This made me think of my research question. Is AIDS as the cause of death even correct? Being in HIV care for almost since the beginning of the pandemic, we (in an HIV clinic) heard many times that when a patient died, the coroners didn’t feel to do an autopsy and just assumed the patient died of AIDS. However, in many instances, AIDS is not necessarily the cause of death. I hypnotize that some of the death certificates with AIDS as the cause of death are wrong. The study design could include autopsies of all people whose cause of death is presumably AIDS. Maybe, the autopsy should be done from two different medical examiners/coroners because of the bias of people who have an AIDS diagnosis. Because in many cases funding is scarce, as an alternative verbal autopsies could be performed. In that case, I would not only include interviewing families/friends but also clinician (provider, nurses, social workers) that might shed a different light on the cause of death. Not each death is AIDS related.
2. Arcaya’s study prompted me to think of other outcomes are associates with proximate foreclosures of properties. My research question would be, “Is proximate foreclosure of properties associated with depression. I hypnotize that people who are threatened by property disclosure may experience a stressful event that could lead to depression. The study design could include a self-report instrument such as the PHQ-9. However, this might not be the best predictor of depression because we can’t compare the PHQ-9 outcome before those individuals heard about the foreclosure of their property. Sure, we could ask them to report their symptoms before they were exposed to the threat but will that be unbiased? Maybe, a more effective measure would be a biomarker such as looking at cortisol which is elevated in the event of stress. Maybe, this could be done several times throughout the study of those individuals and track if there is an increase. We of course wouldn’t have a cortisol level on someone before the exposure of the threat but throughout the stressful event. The study might show evidence of depression and further studies could look into overall mental illness due to foreclosures of properties and how much of a burden that is.
3. The Banks article was another interesting article. At the HIV clinic where I work as a part-time research nurse, I have conducted a clinical study for HIV-positives who are 50 years and older. Instead of looking at the SES and education as predictors, we could look at the social provisions of HIV-positive aging populations. The research question could be: does loneliness in HIV-+ aging populations predict morbidity. We could hypnotize that HIV-+ aging populations with higher degrees of loneliness experience higher morbidity. There a few good instruments to measure loneliness such the Social Provisions Scale by Russell and the UCLA loneliness scale (we use both in our clinic). In Banks' study, the researchers used self-reported morbidity outcome. In my case here, we could use the Veterans Aging Cohort Study (VACS) index as an outcome. The VACS index includes biomarkers such as age, sex, race, CD4 count, HIV viral load, ALT, AST, platelets, fib-4 (liver marker), eGFR, hep. C status (we calculate this index for our patients). The VACS index is a great predictor of morbidity and mortality and calculates 1- and 5-year mortality. We would have patients completed those loneliness surveys and the electronic health record calculates the VACS index. A relatively simple study to determine if loneliness predicts the VACS index as a health outcome. Banks compared the US to England; we could do the same compare HIV-+ aging populations in the US to those from England or anywhere else.

Have a nice weekend.

1)  Data Source: NHS (Nurses Health Study)

 Weak Research Question:  The article by Lee addressed the question of the impact on caregiving on CHD. The problem with this particular data source and their research question is that they did not have access to SES or some sort of a financial burden measure.   This may partially explain the association found between caregiving for spouses and CHD, but not with caregiving and CHD with children or parents. If the spouse cannot work, there is more financial burden and thus more stress.  Additionally, the NHS data source lacked information on the particular illness or disability, the duration of caregiving, the health status prior to giving care, or which care recipients provided the most stress or reward. All of this would be limiting to any research question looking for the impact of caregiving on a health outcome.

Alternative Data Source for Weak Research Question:

An alternative data source to SES might be zipcode or another measure estimating SES in the NHS. Or perhaps doing this study in another data source that has all the same measures as NHS but also has some SES measures (not sure if this exists).

Stronger Research Question:  Do stress levels due to financial burden, caregiving, and lack of time increase the risk for CHD?

-----------------------------------------

2)  Data Source: Death Certificates

Weak Research Question:  Did the introduction and rollout of super-sized sugar sweetened beverages in the years 19XX-19XX increase CHD deaths as compared to prior years? This is a weak research question when using the National Center for Health Statistics “cause of death” reported on death statistics due to the  extreme inaccuracies of Cardiovascular Deaths reported on death certificates.

Alternative Data Source for Weak Research Question: An alternative data source for a question like this might be to use the Kaiser database system which has doctors record the reason for death.

Example of a stronger Research Question would include something you want to learn about related to all-cause mortality rates. For example has the ratio of all-cause mortality in women compared to men increased in 2014 compared to 1965?

ARIC

Strong: Communities with more extreme temperature fluctuations will have steeper increases in systolic blood pressure over time, but otherwise better results on a host of heart health indicators. ARIC has these measures taken in 4 quite different communities. Temperature data is easily collected from government agencies.

Weak: High levels of HDL cholesterol (≥60 mg/dL) are associated with better cognitive functioning in older adults, and secondarily this relationship is modified by lead exposure. The Baltimore Memory study is better equipped to test this hypothesis with longitudinal collection of blood (could measure HDL and lead) and a battery of cognitive tests.

Death Certificate

Strong: Counties that produce less meat will have lower death rates overall and deaths at older ages than counties that produce more meat. Death certificates in the U.S. are good at capturing deaths, but perhaps not so good at capturing specific types of death as we learned in the Pagitipati article. Meat production data could be retrieved from the USDA.

Weak: Death rates in Burma among children are associated with acres of opium in the 10km surrounding their homes. Most developing countries do not have death certificates. Verbal autopsies in randomly selected regions in Burma could help answer this question.

HRS

Strong: air pollution is associated with reduced levels of exercise in older adults and this relationship helps explain part of the association between air pollution and cognitive function in older adults. Many studies have assessed how air pollution affects lung function and health during exercise, but few (any?) have assessed whether air pollution is associated with reduced amounts of exercise. Air pollution data is available from monitors and modeling.

Weak: living close to a fracking site is associated with insomnia, sleep quality, and sleep disturbances. HRS does not have enough participants living near a site to conduct this research. Electronic health record data from the Geisinger Health System and hospitals in Pittsburgh could be used to assess associations with diagnosed insomnia. Data on sleep quality and sleep disturbances would need to be collected through a new study.

CA’s Office of Statewide Health Planning and Development (OSHPD) Patient Hospital Discharge Data

This is a comprehensive data source for every inpatient hospital discharge at a licensed general acute care hospital in California (excluding the VA). The reported data includes patient demographic information (e.g. age, sex, county of residence, and race/ethnicity), diagnostic information (discharges using ICD9 codes), treatment information, total charges and expected source of payment. Available data dates back to 1997, and data is collected yearly.

The OSHPD data is useful for looking at disease incidence (particularly rare diseases) at the state level (or county level) for a particular year or within a certain time-span since the discharge code(s) can easily provide this information. For example, we can use the OSHPD data to look at changes in incidence of salmonella infections and costs associated with these infections from 2005-2010 in Los Angeles County. One problem with using OSHPD data is hospital readmissions are counted as unique discharges. Therefore, using counts of hospital discharges may not accurately represent the true number of cases.

It is difficult to make sound causal interpretations of studies using this data source. For example, a question such as “what the association between type 2 diabetes and myocardial infarction” would be very difficult to answer. Information on patient’s current health status would not be available, and there may be unmeasured confounding factors that can affect this association. A better data source would be from a healthcare provider like Kaiser where we have comprehensive medical history of each patient.

Multi-Ethnic Study of Atherosclerosis (MESA)

MESA is a longitudinal study (2000-2011) on the characteristics and risk factors of subclinical cardiovascular disease among different races/ethnicities (African Americans, Whites, Chinese-Americans, and Hispanics). A variety of data was collected, including biological, sociodemographic, life habits, psychosocial, etc.

The unique features of MESA’s data allow us to examine associations between race/ethnicity, risk factors, and cardiovascular outcomes. One possible question is: How does race/ethnicity modify the association between kidney disease and progression to incident cardiovascular disease? Comprehensive baseline data on cardiovascular and kidney health is available, and using this information we can also determine participants who were free of prevalent CVD at baseline. The longitudinal nature of the data follows these participants over time until their first CVD event, if it occurs at all, and using this information we can determine if the time to event differs between races/ethnicities.

All of MESA’s participants were middle-age or older at time of enrollment (age ranges from 45-84 years). Therefore, questions about exposures earlier in life typically cannot be answered using MESA. For example, a question such as “Does binge drinking among young adults increase risk of cardiovascular disease later in life” requires information on exposures during youth. This information may not be available in MESA, and if it is it would be based on participant recall, which is prone to recall bias. An appropriate study for this would be the Coronary Artery Risk Development in Young Adults (CARDIA) Study. This study enrolled young adults (ages 18-24 and 25-30) and aims to follow them for 30 years.

Study of Osteoporotic Fractures

SOF is a longitudinal study (16 years) that originally focused on risk factors and falls among non-Hispanic white women over 65 years of age (African American women were later also enrolled into the study), but has expanded to explore the process of aging in older women. Data collected includes medical history, anthropometric measures, cognitive and physical function, biologic measures, etc.

As this is a study consisting of older women, the data available makes it especially appropriate for answering questions related to aging. For example, SOF is a good data source to explore the question “What is the association between changes in physical function and cognitive decline?” SOF data includes longitudinal measures of both physical and cognitive functions, along with many other variables that can be possible confounders in this association (e.g. heart disease, diabetes, depression).

However, SOF cannot be used to explore gender differences. For example, “Does decline in physical function affect cognitive functioning differently in men and women?” would not be an appropriate question for SOF since the study only consists of women. One possible alternative is the Sacramento Area Latino Study on Aging (SALSA). In this study, Mexican-American men and women were followed yearly for ten years and underwent clinical, cognitive, and functional assessments at each follow-up. With SALSA, we can examine gender differences in the association between changes in physical function and cognitive decline. Since assessments were conducted yearly, incremental changes in both physical and cognitive function were more likely to be captured.  It is worth noting, however, that the results may not be generalizable since the study consists of only Hispanic older adults. 

Data Source = ARIC

Strong research question:

The ARIC study is particular well equipped to investigate a question such as: Do biological changes (arterial stiffening, changes in lipid levels, and homeostasis activity) mediate the effect of known risk factors on the risk of CHD events controlling for potential confounders? I make this judgement based on the substantial detail on biometric measurements that were taken at both baseline and 3 years later. Having these two time points would allow for mediation analyses for known risk factors that were also collected and that are updated via the annual telephone interviews. It would be a stronger data set if biometric measurements were taken more often (i.e. annually) but I believe mediation analyses could still be done with just these two time points especially for risk factors that changed between baseline and the 3 year follow up exam.

Weak research question:

Although the study population comes from four different communities in the U.S. that are representative of their respective community populations, I do not believe the ARIC study data alone would be well equipped to answer questions about community level exposures such as: What is the effect of community level environmental exposures (i.e. walkability, pollution, or crime rates) on the risk of CHD events. These types of community level exposures were simply not measured in the ARIC study; however the data for such variables could be found in other data sources such as data from pollution monitoring stations run by the the EPA or crime rate data from the census and Bureau of Justice Statistics. These data could then be linked back to participants’ communities in the ARIC study, given the participants’ address of residence to identify the participants’ census tract of residence via geocoding.

Data Source = US National Center for Health Statistic birth and death certificate data

Strong research question:

The data source described in the Pickett et al. article that linked birth and death certificates for all infants born in the US and died in the first year of life would be particularly strong to assess the question: What is the effect of mandatory folic acid fortification of enriched grains on infant mortality in the first year of life? The mandatory fortification of enriched cereal grain products was begun in 1996 and fully implemented by 1998 which is in the same time frame as this data set. This data set likely contains data for all years even past 1998 so for this particular question we may be interested in the 1989 to 1991 data (prior to fortification) compared to the 1998 to 2000 data (post fortification). The effect that fortification had on incidence of neural tube defects has been well studied but less research has been done on whether infant mortality in the first year has been influenced by fortification since folic acid has been shown to be beneficial to overall fetal development not just for neural tube development. Additionally, the data available on mothers (region of residence at time of birth, mother’s age, nativity, marital status, etc) that were used in the Pickett et al. study (to evaluate the association between the Back to Sleep campaign on social inequalities) seem to also be all potential confounders for the association between folic acid levels and infant mortality.

Weak research question:

Unfortunately this data set does not have food frequency questionnaire (ffq) data on the mothers in the data set and thus the data set would not be equipped to answer a question such as: What is the effect of folic acid levels or folic acid intake on the risk of infant mortality in the first year of life? This would require individual level data about folic acid that could be obtained through ffq or blood samples of the mothers. An alternative dataset that could be used to answer this question is the National Birth Defects Prevention Study which does take blood samples of the mothers and allows for measurement of mothers’ folic acid levels as well as a ffq. Potentially data from the NHANES data sets could also be used but would need to be linked to the outcome data in some way. Unless the study aims to evaluate a population-based infant mortality rate as the outcome and uses the NHANES data to calculate an average folic acid level for a certain community for which the rate is being calculated. Thus the question would be whether community level folic acid levels are associated with that community’s infant mortality rate in the first year of life, if the infant’s family is unlikely to have moved out of the area that the mother resided in during her pregnancy.

Data Source = Nurses’ Health Study (NHS)

Strong research question:
A potentially strong research question that could be addressed using the NHS would be: Does physical activity influence the risk of colorectal cancer? Since this cohort is being followed up every two years, their levels of physical activity could be measured as a time-varying exposure and the long follow up time would be a strength for evaluating a late-onset disease like colorectal cancer. Also important covariates are accounted for in this data set such as smoking, alcohol intake, BMI, saturated fat intake, and I believe additional dietary information such as red meat intake is also available in the NHS.

Weak research question:
Since this is a prospective cohort study, one type of research question it may not be well equipped to investigate would be those about rare diseases, such as ovarian cancer. Therefore an example of a weak research question would be: Is postmenopausal hormone use associated with incidence of ovarian cancer? The issue is that lifetime risk of ovarian cancer is about 1.33% and thus only about 1% of all the nurses in the cohort would be expected to experience the outcome, and although the cohort is very large, it is likely to still be underpowered to evaluate this question. Other factors to consider are that BRCA1 and BRCA2 genes are strong risk factors for developing ovarian cancer and I believe in the original NHS, germline genetic data was not collected on the entire cohort. However, this genetic information may not be a confounder in the evaluation of the association between hormone use and ovarian cancer unless we believe that there could be a non-causal pathway between hormone use and ovarian cancer that involves the woman’s family history of hormone use (i.e.mother used hormone therapy and encouraged daughter to use it as well), which would also be associated with inheriting a BRCA mutation. Alternative data sources for studying the effect of postmenopausal hormone use on ovarian cancer could come from the Danish national datasets that could allow for a case-control study of all ovarian cancer cases in Denmark, where hormone use would also be readily available along with numerous other covariates.

I want to try to answer the following research question using 3 data sources: Does obesity or physical inactivity in U.S. pre-teenage children (age 6-12) predict cardiovascular morbidity/mortality in adulthood?

Atherosclerosis Risk in Communities Study

Strengths of ARIC: Physiological measures (e.g., EKG, ultrasound, spirometry, blood sample), hospital records, interview, death records

Weaknesses of ARIC: Don’t explicitly ask about childhood history, regional differences in diet/exercise (e.g., no recruitment from a US West coast city)

Alternative study design: No need to abandon the ARIC study completely for this research question – Add a site on the US West coast, add questions about childhood history, obtain pediatrician’s records if possible, obtain school records if possible (e.g., sports, yearbook)

National Vital Statistics System Death Certificate Data

Strengths: Since the US is a developed country, there is an existing system to classify causes of death.

Weaknesses: Inaccuracies of death certificate diagnoses when converting to ICD codes, Overrepresentation/Underrepresentation of diagnostic categories

Alternative study design: No need to abandon the National Vital Statistics Systemcompletely for this research question – Obtain autopsy data from local hospitals, Possible verbal autopsy method if formal autopsy not available

Nurses’ Health Study

Strengths: Physical activity and cardiovascular measures are obtained

Weaknesses: Don’t explicitly ask about childhood history, Men not included

Alternative study design: No need to abandon the NHC study completely for this research question – Add questions about childhood history, obtain pediatrician’s records if possible, obtain school records if possible (e.g., sports, yearbook)