Question 1
In an observational study, the investigator “watches” as subjects themselves choose which group they will be in (exposed or unexposed); in an experimental study, the investigator assigns participants to their exposure group.
Both the exposures and outcomes have already occurred at the start of a retrospective cohort study; the outcomes have not yet developed at the start of a prospective cohort study. Thus, a retrospective cohort study investigates prior outcomes and a prospective cohort study investigates future ones.
A cohort study defines subjects according to their exposure level and follows them for disease occurrence; a case-control study defines cases of disease and controls and compares their exposure histories.
Question 2
The main similarity is both compare two or more exposure groups, which are followed to monitor outcome rates. The main difference is that the investigators allocate the exposure in experimental studies, and the participants choose their exposures in cohort studies.
Question 3
D. A retrospective cohort study
Question 4
a. This is a community randomized trial (experimental study) because the villages were randomly assigned to the intervention or the control group.
b. This is a cross-sectional study because the exposure and outcome are being assessed at one time.
c. This is a case-control study because exposures are being compared in two groups defined according to disease status.
d. This is a prospective (birth) cohort study because the babies are identified in the present (i.e., at the time of study initiation) and followed into the future to determine both exposure and outcome status.
e. This is a case-control study because exposures are being compared in two groups selected according to the presence or absence of the outcome of interest.
f. This is a retrospective cohort study. It’s a cohort study because the subjects were selected by exposure status. The exposure was the first birth outcome and outcome was the subsequent birth. The study is retrospective because medical records were used to reconstruct a cohort and because the exposure and outcome had already occurred at the time the study was initiated.
g. This is a cross-sectional study because exposure and outcome status are assessed once and at the same time for each participant and no follow-up was done.
h. This is a (nested) case-control study. First, it is a case-control study because exposures are being compared in two groups defined by the outcome and, second, it is nested because both cases and controls are selected from participants in a cohort study.
i. This is a randomized controlled trial (experimental study) because the subjects are randomly assigned to one of two intervention groups.
j. This is a case-control study because exposures are being compared in two groups defined according to disease status.
Question 5
Yes
No
No
Yes
Question 6
False
False
False
False
False
False
Question 7
True
False
True
Question 8
In individual trials, the treatment is allocated to particular people; in community trials, the treatment is allocated to entire communities.
Question 9
Preventive trials investigate measures that stop or delay the onset of disease; therapeutic trials investigate measures that treat existing disease.
Question 10
In parallel trials, the treatment is administered such that individuals in each group simultaneously receive only one study treatment; in crossover trials, two or more study treatments are administered one after another to each group (such that participants receive all interventions and only the order of treatments differ).
Question 11
Cluster randomized design
Prospective cohort
Cross-over
Factorial randomized trials
POR (prevalence odds ratio)
RR
OR
SMR
Because data were collected prospectively, the most appropriate measure is a risk ratio or incidence rate ratio. Because birth defects are most often measured using prevalence rather than incidence, answer A could also be accepted.
Masking is the process of preventing certain groups in a study from knowing who receives the intervention. It is important to mask to eliminate the possibility that knowledge of treatment will influence our outcome. Not all studies can use masking.
A double-blinded study is typically describing a study in which both participants and investigators are masked.
When randomization is successful, both groups in a treatment arm should be statistically exchangeable with respect to the attributes we suspect may confound the relationship between our exposure and outcome (in this case, age and outcome of previous pregnancies). This statement therefore suggests that the randomization was successful because groups are similar with respect to known confounders, and therefore was likely successful with respect to unknown confounders as well.
Efficacy: The extent to which a drug/intervention has the ability to bring about its intended effect under ideal conditions.
Effectiveness: The extent to which a drug/intervention achieves its intended effect under usual conditions.
The authors attempt to estimate both effectiveness and efficacy. The analysis including all women regardless of whether or not they took folate as prescribed approximates effectiveness, and the analysis including only women who complied with their treatment arm approximates efficacy.
Equipoise is a state of genuine uncertainty about the benefits (or harms) that may result from a treatment or intervention. In this case, it was ethical to stop the trial after one year because the results are sufficiently persuasive of folate’s benefit that we are no longer uncertain about its benefits.
Question 12
Phase 4
Phase 1
Phase 3
Phase 2
Question 13
Eligibility criteria such as these are based upon scientific, safety, and practical considerations.
Practical - intend the intervention to be used by this specific population
Safety/Scientific - must consider the safety of the fetus in the context of the known scientific properties/mechanism of the drug
Unethical to assign women with immediate medical need to the placebo group.
Question 14
A process by which the investigator assigns subjects to the treatment and comparison groups. Subjects have an equal chance of being assigned to either the treatment or comparison group.
The group to whom the study results may be applied or generalized.
The study subject who doesn’t know whether he or she is in the treatment or comparison group.
The run-in period occurs before enrollment and randomization to determine which participants are able to comply with the study regimen. Potential participants are placed on the test or control treatment for a certain period of time to assess their tolerance to and acceptance of the treatment and to obtain information on compliance
A placebo is an inactive substance, such as a sugar-coated pill, that is given as a substitute for an active substance. A sham is a bogus procedure that is designed to resemble a legitimate one. Both placebos and sham procedures permit the study participants and caregivers to be masked.
Equipoise is a state of mind characterized by genuine uncertainty about the appropriate course of action - that is, to give or withhold a particular treatment. It is ethical to conduct an experimental study only when there is a state of equipoise in the expert community.
Question 15
The anticipated difference between the treatment and comparison groups
The background rate of the outcome
The probability of making statistical errors known as alpha (type I error, or false positives) and beta errors (type II error, or false positive)
Question 16
The anticipated difference between the treatment and comparison groups
The background rate of the outcome
The probability of making statistical errors known as alpha (type I error, or false positives) and beta errors (type II error, or false positive).
Question 17
Informed consent is an educational and decision-making process in which potential volunteers must be effectively informed of the purpose, procedures, and possible risks and benefits associated with participation in a study.
Question 18
The purpose of an IRB is to help ensure ethical research practices. IRBs are typically composed of a least five members from diverse backgrounds including at least one nonscientist. IRBs review research applications from investigators in order to ensure that the application demonstrates all of the necessary provisions for human subject provisions.
Question 19
Unique features include random assignment of subjects to the treatment and comparison groups to control for confounding and to reduce biased allocation, and the use of placebo controls to permit masked assessment of the outcomes.
Question 20
a. False, not all treatment groups are compared with a placebo; they may be compared with another treatment. It would be unethical to give participants in the comparison group a placebo when a safe and effective treatment is already available.
b. True. (But this isn’t the only use of blinding.)
c. True (it prevents an unbalanced distribution of participants to the treatment groups from occurring by chance).
d. False, an intention-to-treat analysis obtains a measure of effectiveness, not efficacy.
e. False, each participant gets one treatment or both treatments together, or the placebo.
f. True
g. False, the randomization process may not result in a fully balanced baseline.
h. False, prevention trials almost always have rarer outcomes (preventions) relative to treatment trials, where ill people are treated and cures are likely to be more numerous. The rarity of the prevention outcome means a larger number of participants are required
i. True
j. False, cancer takes a long time to develop so a randomized trial would not be an efficient way to investigate this relationship (and there are ethical issues about deliberately exposing people to possible carcinogens, even if the carcinogenic potential were unknown). A more efficient and ethical design would be an observational study.
k. False, the goal of randomization is to ensure that confounding and selection factors are equally balanced between groups. Equipoise is about ensuring benefits potentially outweigh risks.
l. False, large overall study size in itself does not ensure equal sample sizes in the groups.
Question 21
a.
This study is a cluster randomized trial. The unit of randomization is the community (n=22) and communities were then pair-matched on incidence of child diarrhea. Treatment allocation was randomized during a public event through a lottery/ball-drawing – note how well described the randomization process is! The randomization sequence was created beforehand by an impartial witness.
b.
The main purpose of randomization is to assure that characteristics of participants are as similar as possible across groups at the start of the study – i.e., to control for confounding, both measured and unmeasured. The main worry with cluster randomization is that there will be “contamination” across the clusters; that is, the concern that the outcome of participants who were not randomized to “treatment” will be influenced by the treatment assignments of those participants who were randomized to treatment due to geographic or social proximity to participants randomized to treatment. This will bias the results because some members of the control (or comparison) group are now exposed to treatment.
c.
If compliance/adherence is low, then there is potential for bias (in particular, selection bias – we will cover this in detail later in the course). People who drop out of studies or who do not comply with treatment are usually different, and potentially have different outcomes, than those who remain in the study and have good compliance. This systematic difference is what leads to selection bias. Using intention-to-treat (ITT) analysis in situations where participants have deviated from their assigned treatment means that the treatment and control groups will look more like each other – i.e., you are less likely to observe an effect of treatment. This means that ITT analysis is “conservative” with respect to potential selection bias (versus a per-protocol/as-treated analysis, which would include in the treatment group only those individuals who actually received treatment, and thus potentially introduce bias). In this case, compliance was particularly low. You could choose either method as long as you justify your analysis.
d.
The authors explain that SODIS has been shown to be highly efficacious in laboratory settings. SODIS has also seemed to be effective in previous randomized controlled trials, but the authors note that these previous trials were conducted under highly controlled, unusual settings that enforced high compliance rates. Therefore, such trials might be better described as providing evidence for efficacy, since they did not have normal field conditions in which to assess compliance rates. In the Mausezahl study, field conditions were much more typical of the kind of community the intervention is normally deployed in. In this setting, compliance rates were quite low, which meant that the effectiveness of SODIS was also low – in fact they did not find a statistically significant effect of the intervention.
e.
Blinding participants and assessors in this trial would be extremely difficult, if not impossible, as noted by the authors (see pg. 8). The intervention does not lend itself to blinding: the treatment requires active, knowledgeable participation by the families (i.e., leaving plastic water bottles out in the sun), and whether they are participating is visible to all observers, including the community field workers who are assessing compliance. However, this was inherent in the design of the study, and the geographic distribution of the communities helped ensure no cross-contamination.
Question 22
There are several possible answers. For example, design a simple protocol, enroll motivated and knowledgeable participants, exclude subjects who might have difficulty complying, present a realistic picture of the required tasks at enrollment, maintain frequent contact with study subjects, and use items such as pill packs to make it easier to comply with the treatment regimen.
Randomize a large number of study subjects.
Mask investigators to group assignment.
Question 23
Primary outcomes are the main condition that the trial has been designed to evaluate and the outcome used to power the study (or used for sample size calculations); secondary outcomes are the remaining end points and are considered less important.
Question 24
Loss to follow-up decreases the number of individuals who can be included in the analysis and so reduces the statistical power of the study. Also, if those who are lost have different rates of disease than those who remain, the study results may be biased.
Question 25
Investigators could review the participant’s medical records to ensure the patient was accurately reporting their last blood pressure reading.
Investigators may also have asked the parents or primary caregivers of the children about the side effects their children experienced to corroborate the outcomes.
Medical claims data could be used to validate the reporting of events by the clinicians.
Question 26
No lines were drawn between the exposure (cotton-fiber endotoxin) and the other risk factors for liver cancer because it is unlikely that endotoxin exposure is associated with these other risk factors. This particular DAG explicitly assumes that these other risk factors are neither confounders nor part of the causal pathway between cotton-fiber endotoxin and liver cancer. The paper does mention these risk factors as potential, though unlikely, confounders; you could have erred on the conservative side and included more arrows given this.
This is an example of a primary study base (the study is nested within a defined cohort). When you have a primary study base, it can be hard to identify all of the cases in the defined population. (Secondary study bases start with the cases and then have a hard time finding appropriate controls.) The authors went through the tedious process of looking through two tumor registry systems, cross-checking, and then examined any reports that were discordant in the systems. The authors seem to be pretty thorough, so we’ll say case ascertainment was reasonably complete.
Note: One advantage of a primary study base is that it can be enumerated.
Generally, case-cohort designs can be used to calculate ORs that estimate the CIR. However, in this paper, the authors estimated hazard ratios (HR) in lieu of ORs. (Don’t worry about HRs now, we’ll talk more about them later.) First they reconstructed the person-time contributed by the sub-cohort by assessing when each person was no longer at risk (due to leaving the factory, diagnosis of liver cancer, or death by competing risk) and by measuring exposed and unexposed periods while working. Second, they had complete information on basic demographics at baseline, including age, so used this sampling fraction to weight the person-time contributed by the sub-cohort to reflect what would have been measured if they followed the full cohort. This enables calculation of the incidence rate. This analysis is beyond the level of complexity we are discussing for the case control design, but you may see it in case-cohort papers because these studies typically take place in the context of a defined cohort used for other purposes.
The investigators treated them as cases only, removing them from the control group. The correct approach is to include them as both cases and controls since they contributed person time to both groups.
Question 27
A.
Nested case-control using density sampling
B.
This is a non-concurrent study. The investigators retrieved the serum samples to assess past exposure after the outcome was diagnosed, that is, after the exposure and outcome had happened (the methods section makes it clear that they collected cases of breast cancer at one time – 1998 – instead of collecting cases as they occurred).
Exposure information was collected prospectively via the collection of blood samples from the cohort from 1959-1967, before outcomes were recorded. The assays for DDT were run in 2000-2001, but because the serum had been collected prior to the outcomes, we would still consider this prospective exposure assessment.
C.
According to the abstract, controls were selected from the underlying cohort (members of the Child Health & Development Studies) and matched to cases on birth year. We can assume from the study design that they used risk-set sampling. If you ventured into the methods section you also learned controls were selected at random and were required to be disease-free at time of the case’s diagnosis.
D.
Risk-set sampling helps ensure that the exposure distribution among the controls is the same as it is in the person-time in the study base for the cases, and thus allows the authors to calculate a (reasonably) unbiased estimate of the IDR.
E.
The incidence density ratio (IDR).
F.
The control sampling fractions (‘r’ in Jen’s lecture) for exposed and unexposed controls must be equivalent to the corresponding exposure distributions within the PT of the study base for the cases. In other words, the ratio of unexposed controls to unexposed PT equals the ratio of exposed controls to exposed PT. See Jen’s lecture, lab slides, or required reading (especially ME3 Ch. 8) for a more comprehensive explanation and equations if this is confusing.
Question 28
This problem is based on an actual study that was conducted, but there are probably many different case-control studies one could design for this question. In the actual study:
Case-HCWs had a documented occupational percutaneous exposure to HIV-infected blood, HIV seroconversion temporally associated with the exposure, and no other concurrent exposure to HIV.
Control-HCWs had a documented occupational percutaneous exposure to HIV-infected blood and were HIV seronegative at the time of exposure and at least 6 months later. Cases were selected from the national occupational surveillance system, while controls were selected from a “passive surveillance project maintained by the CDC” that included data from ~300 healthcare institutions.
Sampling techniques: For cases, you might decide to include all available cases, or a random sample. For controls, you could randomly select up to 4 controls from the healthcare institution surveillance system for each case. It is unclear from the actual study whether the investigators matched cases and controls on time of exposure; however, this would be an essential strategy if you were doing a density sampled case-control study.
To determine if this study might be a primary or secondary study base, the key question to consider is whether you can define the population of interest without reference to the cases. In this case, a plausible population of interest is all health-care workers in the United States who were exposed to percutaneous HIV-infected blood on the job and active in 1995 (or some period such as 1990 – 1995, 1980 – 1995). Since we can define this population in a way that does not mention cases, this is a primary study base. The questions are then how to sample from this base and how to identify cases. In the actual study, the investigators would have to assume that the controls/data used from the healthcare surveillance system are representative of all HCWs (i.e., they represent the source population for the cases). It might be possible to compare their demographics to data collected from HCWs (i.e., through professional associations or ongoing surveys of HCWs like the Nurses’ Health Study) to test this assumption. In terms of cases, we are fortunate to have a complete registry of cases in this instance through the national surveillance system.
Sampling from this group could be challenging, but luckily CDC’s passive surveillance system includes information on occupational incidents such as needle pricks.
Question 29
A.
Exposure = fan use during sleep
Outcome = sudden infant death syndrome (SIDS)
B.
This is a case-control study. The authors probably chose this study design because SIDS is a relatively uncommon outcome (0.53 deaths/1,000 live births). It would have been time-consuming and expensive to enroll and follow a prospective cohort of infants that was large enough to eventually include enough cases of SIDS for the analysis. Thus, a case control study is advantageous because the investigators can start by identifying cases of SIDS rather than waiting for them to occur.
C.
This is a primary study base. The authors identified the population first (11 California counties) between May 1, 1997 and April 30, 2000; and then identified cases based on that population. This means that controls are selected from this population also. The advantages are that we can be more confident that the controls come from the population that gave rise to the cases (i.e. because they are all the other births in that county that would have become cases had they died of SIDS). Also, we can be more confident that exposure in controls estimates the distribution of disease in the general population. Disadvantages include the following: we must consider whether all cases have been identified, controls from this population may be less motivated to participate, it may be expensive to track down and interview controls from such a wide population base, and recall bias may be more of a concern.
D.
Yes. They matched on county, maternal race/ethnicity, and infant age. Matching allowed the investigators to make sure that their cases and controls were similar on those factors. This is important if they suspected that these factors will have a strong influence on the relationship between the exposure and outcome. However, matching on these factors means that they weren’t able to analyze the association of the matching factors in relationship to the outcome.
Question 30
It is desirable to conduct a case-control study when the exposure data are difficult or expensive to obtain, when the disease is rare, when the disease has a long induction and latent period, when little is known about the disease, and when the underlying population is dynamic.
Question 31
Advantages: They usually come from the same source population as the cases and so they are likely to be comparable.
Disadvantages: They are time-consuming and expensive to identify, they are usually not as cooperative as hospital controls, and their recall of prior exposures may not be as accurate as that of cases.
Question 32
Advantages: Case-control studies take less time and money to conduct than cohort and experimental studies, they are well suited for studying rare diseases and diseases with long induction and latent periods, and they can provide information on a large number of possible risk factors.
Disadvantages: The possibility of bias is increased, and it may be difficult to establish the correct temporal relationship between the exposure and disease because the data are retrospective.
Question 33
An open or dynamic cohort is conducted in a population defined by a changeable characteristic such as residence in a specific place. Thus, its members come and go, depending on whether they have the characteristic.
A fixed cohort is defined by an irrevocable event, so it does not gain any new members.
A retrospective cohort study looks back in time and examines exposures and outcomes that have already occurred by the time the investigator begins the study.
A prospective cohort study looks forward in time and examines future outcomes in relation to past or current exposures.
A standardized mortality ratio (SMR) study is a special type of cohort study in which the mortality experience of an exposed group is compared with that of the general population. It is commonly conducted in occupational settings.
Question 34
The cohort consists of physicians in the UK: 40,701 total; 34,494 men and 6,207 women. The cohort is closed since no new members are added over time. The cohort is fixed.
Smoking was the exposure, measured by the daily amount of grams of tobacco smoked per day (1 gram = 1 cigarette) at the time of inquiry (baseline). They also used other exposure classifications: dichotomized (non-smokers vs. current? smoker), length of time since quitting among ex-smokers, and method of tobacco use.
Because the investigators measured smoking at the time of inquiry and not lifetime smoking, “any gradient [they] may observe in relation to the amount of smoking will be an understatement of the true relationship.” (p. 5001) For example, a light smoker may have been a heavy smoker earlier in their life and similarly, a heavy smoker may become a light smoker. By ignoring the cumulative exposure to cigarettes, the resulting groups may actually be very similar in terms of their exposure, potentially washing out some of the true relationship.
This study examined several outcomes, all related to mortality. The primary outcome of concern was lung cancer mortality. Deaths were obtained through a countrywide death index (Registrars-General). Supplemental information was provided through the "General Medical Council" and the "British Medical Association". All deaths were classified according to cause. Detailed information on the 88 cases of lung cancer was obtained by confirmation from the doctor who certified the death. Specific criteria are outlined in Table IV.
Incidence rate ratio (IRR)
The study identifies these as standardized death rates per annum per 1000 men. Incidence rate of lung cancer among smokers: 0.90 / 1000 person-years
Incidence rate of lung cancer among non-smokers: 0.07 / 1000 person-years
Incidence rate ratio: 0.90/0.07 = 12.86
Among approximately 40,000 British medical professionals between 1951 and 1956, the rate of death (incidence density of death) from lung cancer in smokers was 12.86 times the rate of death from lung cancer in non-smokers.
Question 35
Non-concurrent cohort
Many answers possible. Among the most common, any three of:
Can ensure temporality
Can capture rates
Can calculate many measures of association
Good for single, rare exposure (1% exposure prevalence in this case)
Possible to assess multiple outcomes
Non-concurrent study design reduces data collection costs and time.
Some outcomes (e.g. ADHD) may be fairly common which is well suited to a cohort design
Prospective exposure assessment ensures that any misclassification of exposure should not be differential by disease status because exposure is recorded before the outcome is known. Non-differential misclassification of exposure is still a concern. Remember, non-differential misclassification is still information bias.
Question 36
True
False
True
False
Question 37
Study Design: It is a prospective cohort study. (Could also be a retrospective cohort study potentially if they identified the birth cohort later.)
Study Population: The cohort is composed of children born on weekdays between September 1992 and November 1993 who were born to ‘usual residents’ in any of 10 villages in Haryana state in India.
Study question: is there an association between birth weight and blood pressure in childhood among children born in Haryana state in India?
Exposure: weight at birth
Outcome: blood pressure at the time of follow up (7-8 years after birth)
Loss to follow-up due to migration was about 6%. Loss due to death was somewhat larger (8%).
If those who were lost to follow-up differ from those who remained in the study in ways that are correlated with exposure and disease, this will introduce bias in the study. For example, if low birth weight babies are more likely to die or families who are worried about their children’s high blood pressure are more likely to stay in the study, the results will be biased. In this study, the amount of loss-to-follow-up is somewhat concerning (14%) and worth looking into. The researchers will want to do some additional analyses to explore how similar the lost-to-follow up children were to the children remaining in the cohort, and whether or not the loss-to-follow-up introduces bias.
After adjusting for confounders, there was no statistically significant association between birthweight and either systolic blood pressure or diastolic blood pressure among 7-8 year old children in rural India.
Question 38
It is a prospective cohort study. The study population includes 76,240 women from the Nurses' Health Study II who were followed from 1995 to 2005.
The OR of 1.41 should be interpreted as a disease OR: “women who were exposed to DES had 1.41 times the odds of depression compared to women who were not exposed to DES”. In cohort studies, study participants are selected according to exposure status rather than disease/outcome status. In this study, we know which participants do and do not have the exposure and we are calculating the odds of developing disease (depression) given exposure status.
Risk ratio (RR) since the study was prospective and incidence of depression was collected.
When the disease is rare (a typical cutoff is that the total incidence in the population is less than 10%), the OR will approximate the RR.
Note: Sometimes ORs are calculated in place of RRs in a cohort study because analysis is done with logistic regression (which produces ORs).
Question 39
Prospective exposure assessment ensures that any misclassification of exposure should not be differential by disease status because exposure is recorded before the outcome is known. Non-differential misclassification of exposure is still a concern. Remember, non-differential misclassification is still information bias.
Question 40
Of the three types of comparison groups, the internal comparison group comes closest to the counterfactual idea because it comes from the same source population as the exposed group and so is most comparable. However, internal comparison groups are often difficult to identify. The general population is the next best option mainly because it is stable and easy to obtain. Its limitations can include lack of comparability to the exposed group and lack of information on confounders. The comparison cohort is the least preferable option. Although to may be comparable to the exposed group, results from such a study are difficult to interpret because the comparison cohort often has other, possibly noxious exposures
Question 41
Yes
No
No
Yes
Question 42
Poor retention of unexposed individuals without disease to the end of the study
Question 43
Inefficient for rare outcomes
Can be expensive because of the sample size needed and/or follow-up time
May take a long time (prospective cohort studies)
Limited by the availability/quality of baseline exposure data (retrospective cohort studies.
Ethical considerations and/or study procedures may require providing information/education to all cohort members that influences their subsequent exposure status/representativeness
Participants in a study are often influenced by the observation/participation process
Comparability of those enrolled to the general population may be questionable
Loss to follow-up may threaten adequacy of sample size and representativeness. Loss to follow-up may also introduce bias if the reason for loss is associated with the exposure and outcome of interest.
Question 44
Loss to follow-up decreases the number of individuals who can be included in the analysis and so reduces the statistical power of the study. Also, if those who are lost have different rates of disease than those who remain, the study results may be biased.
Question 45
True
False
True
False
Question 46
A.
Study Design: It is a prospective cohort study. (Could also be a retrospective cohort study potentially if they identified the birth cohort later.)
Study Population: The cohort is composed of children born on weekdays between September 1992 and November 1993 who were born to ‘usual residents’ in any of 10 villages in Haryana state in India.
B.
Study question: is there an association between birth weight and blood pressure in childhood among children born in Haryana state in India?
Exposure: weight at birth
Outcome: blood pressure at the time of follow up (7-8 years after birth)
C.
Loss to follow-up due to migration was about 6%. Loss due to death was somewhat larger (8%).
If those who were lost to follow-up differ from those who remained in the study in ways that are correlated with exposure and disease, this will introduce bias in the study. For example, if low birth weight babies are more likely to die or families who are worried about their children’s high blood pressure are more likely to stay in the study, the results will be biased. In this study, the amount of loss-to-follow-up is somewhat concerning (14%) and worth looking into. The researchers will want to do some additional analyses to explore how similar the lost-to-follow up children were to the children remaining in the cohort, and whether or not the loss-to-follow-up introduces bias.
D.
After adjusting for confounders, there was no statistically significant association between birthweight and either systolic blood pressure or diastolic blood pressure among 7-8 year old children in rural India.
Post a Comment