NU.Learning: Nonparametric and Unsupervised Adjustment for Bias and Confounding

Description

NU.Learning forms Local Treatment Differences (LTDs) or Local Rank Correlations (LRCs) within Clusters of experimental units (patients, etc.) who have been relatively well-matched on their baseline X-confounder characteristics. The resulting distribution of LTD/LRC effect-size estimates can be interpreted much like a Bayesian posterior. Yet these distributions have been formed, via Nonparametric and Unsupervised Preprocessing, in purely Objective Ways.

Details

UNSUPERVISED LOCAL TREATMENT DIFFERENCES or LOCAL RANK CORRELATIONS:

Multiple calls to ltdagg(K) or lrcagg(K) for varying numbers of clusters, K, are typically made after first invoking NUcluster() to hierarchically cluster patients in X-space and invoking NUsetup() to specify a numeric y-Outcome variable and a numeric treatment choice or exposure level measure, trex.

UNSUPERVISED INSTRUMENTAL VARIABLES = LOCAL AVERAGE y-OUTCOME EFFECTS:

An OBSERVED Propensity Score (PS) is defined here to be either (i) the local (within-cluster) fraction of experimental units (patients) receiving trex==1 (new) rather than trex==0 (control) or else (ii) a measure of "relative exposure" when the numeric trex measure has (many) more than 2 observed levels. Multiple calls to ivadj(K) for varying numbers of clusters, K, then yield alternative scatters of Local Average Outcomes (LAOs) for Clusters when plotted against their PS estimates and, thus, different possible linear fits or smooth.splines() yielding potentially different inferences about across-cluster Treatment or Exposure Effects.

CONFIRMATION and SENSITIVITY ANALYSES of LOCAL EFFECT-SIZE DISTRIBUTIONS:

For a given value of K = Number of Clusters requested, the output object from ltdagg(K) or lrcagg(K) can be input to confirm() to use (nonparametric) permutation theory to display visual evidence (empirical CDF comparisons) concerning the Question: "Does x-matching Truly Matter?" The NULL hypothesis here is that the x-Covariates used in Clustering / Matching of Experimental Units are actually IGNORABLE. Evidence against this hypothesis is provided when the observed LOCAL Effect-Size Distribution clearly deviates from the purely RANDOM, NULL distribution computed (to any desired precision) by randomly PERMUTING cluster ID labels across experimental units. Furthermore, the statistical significance of differences between the observed and random NULL distributions can be estimated using KSperm(), which simulates the random permutation distribution of the Kolmogorov-Smirnov D-statistic when many tied values occur in both distributions being compared. Finally, the NUcompare() function helps users of NU.Learning decide which Number of Clusters, K, optimizes Variance-Bias trade-offs. Larger values of K tend to yield smaller clusters with better matches and, thus, potentially reduced BIAS. On the other hand, smaller values of K usually yield local effect-size estimates with much lower Variability (higher Precision).

"Most-Like-Me" HISTOGRAMS for DOCTOR-PATIENT discussions of PERSONALIZED MEDICINE:

For a specified vector, xvec, of numerical values of the X-confounder variables used in the current CLUSTERING of eUnits, display histograms of observed LTD or LRC effect-sizes for (i) all available patients and (ii) for the specified number, NN, of "Nearest-Neighbors" in X-confounder space of the TARGET eUnit ...i.e. xvec defines "Me".

Author(s)

Bob Obenchain <wizbob@att.net>

References

McClellan M, McNeil BJ, Newhouse JP. (1994) Does More Intensive Treatment of Myocardial Infarction in the Elderly Reduce Mortality?: Analysis Using Instrumental Variables. JAMA 272: 859-866.

Obenchain RL. (2010) The Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL, Young SS. (2013) Advancing Statistical Thinking in Observational Health Care Research. J. Stat. Theory and Practice, 7: 456-469, doi:10.1080/15598608.2013.772821.

Lopiano KK, Obenchain RL, Young SS. (2014) Fair treatment comparisons in observational research. Statistical Analysis and Data Mining, 7: 376-384, doi:10.1002/sam.11235.

Obenchain RL. NU.Learning-vignette. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

Rosenbaum PR, Rubin RB. (1983) The Central Role of the Propensity Score in Observational Studies for Causal Effects. Biometrika 70: 41-55.

Rosenbaum PR, Rubin RB. (1984) Reducing Bias in Observational Studies Using Subclassification on a Propensity Score. JASA 79: 516-524.

Rubin DB. (1980) Bias reduction using Mahalanobis metric matching. Biometrics 36: 293-298.

Stuart EA. (2010) Matching Methods for Causal Inference: A Review and a Look Forward. Statistical Science 25: 1-21.

Simulate a p-value for the significance of the Kolmogorov-Smirnov D-statistic from confirm().

Description

For a given confirm() output object, KSperm() simulates the NULL distribution of LTDs or LRCs resulting from Purely Random Clusterings of experimental units within the parent data.frame. This NULL distribution is discrete because Local Effect-Size estimates are TIED within-clusters. The observed D-Statistic from confirm() is compared with new NULL order statistics computed by KSperm(), again using stats::ks.test. When KSperm() is called immediately after confirm() and the seed value used in confirm() is known, then both the simulated p-value and the additional NULL KS-D order statistics generated by KSperm() will all be reproducible.

Usage

KSperm(x, reps=100)

Arguments

Details

The observed value of the Kolmogorov-Smirnov D-statistic from confirm() is used here, but its "p.value" from ks.test() is not because it is badly biased downwards. This bias results because the distribution of LTDs or LRCs across clusters is always discrete, due to TIED values within clusters that typically also vary in size. Thus, KSperm() generates "reps" additional, independent, NULL values of KS-D and saves their order statistics. Finally, KSperm() compares the Observed KS-D from confirm() with its simulated NULL order statistics to estimate an appropriately "adjusted" p-value, pv.adj. Note that the simulated pv.adj value estimate cannot be less than 1/(reps).

Value

An output list object of class KSperm:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2010) Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2019) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

confirm, ltdagg and lrcagg.

Internal LocalControlStrategy functions.

Description

Internal functions for the NU.Learning-package. Plot and print functions require the output object from a NU.Learning function as their first argument.

Author(s)

Bob Obenchain <wizbob@att.net>

Hierarchical Clustering of experimental units (such as patients) in X-covariate Space

Description

Form the full, hierarchical clustering tree (dendrogram) for all units (regardless of Treatment/Exposure status) using Mahalonobis distances computed from specified baseline X-covariate characteristics.

Usage

  NUcluster(dframe, xvars, method="ward.D")

Arguments

Details

The first step in applying NU.Learning to data is to hierarchically cluster experimental units in baseline X-covariate space ...thereby creating "Blocks" of relatively well-matched units. NUcluster first calls stats::prcomp() to calculate Mahalanobis distances using standardized and orthogonal Principal Coordinates. NUcluster then uses either the divisive cluster::diana() method or one of seven agglomerative methods from stats::hclust() to compute a dendrogram tree. The hclust function is based on Fortran code contributed to STATLIB by F. Murtagh.

Value

An output list object of class NUcluster, derived from cluster::diana or stats::hclust.

Author(s)

Bob Obenchain <wizbob@att.net>

References

Kaufman L, Rousseeuw PJ. (1990) Finding Groups in Data. An Introduction to Cluster Analysis. New York: John Wiley and Sons.

Kereiakes DJ, Obenchain RL, Barber BL, et al. (2000) Abciximab provides cost effective survival advantage in high volume interventional practice. Am Heart J 140: 603-610.

Murtagh F. (1985) Multidimensional Clustering Algorithms. COMPSTAT Lectures 4.

Obenchain RL. (2010) Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Rubin DB. (1980) Bias reduction using Mahalanobis metric matching. Biometrics 36: 293-298.

See Also

NUsetup, ltdagg and lrcagg.

Examples

data(radon)
xvars  = c("obesity", "over65", "cursmoke")
hclobj = NUcluster(radon, xvars)  # ...using default method = "ward.D"
plot(hclobj)  

Display NU Sensitivity Graphic for help in choice of K = Number of Clusters

Description

This function displays Box-Whisker diagrams that compare Treatment Effect-Size distributions for different values of K = Number of Clusters requested in X-covariate space. After an initial call to NUsetup(), the analyst typically makes multiple calls to either ltdagg() or lrcagg() for different values of K. The analyst then invokes NUcompare() to see how choice of K changes the location, spread and/or skewness of the distribution of Treatment Effect-Size estimates across Clusters. Variance-Bias trade-offs occur as K increases; large values of K may reduce Bias, but they definitely inflate the Variance of LTD and LRC distributions.

Usage

  NUcompare(envir)

Arguments

Details

The third phase of NU.Learning is called EXPLORE and uses graphical Sensitivity Analyses to show how Treatment Effect-Size distributions change with choice of NU parameter settings. Choice of K = Number of Clusters requested is guided, primarily, by NUcompare() graphics. Equally important are the analyst's choices of (i) which [and how many] of the available baseline X-covariates to "adjust for" and (ii) which clustering algorithm and dissimilarity metric to use. Unfortunately, changing these latter choices requires the analyst to essentially "start over" ...i.e. invoking NUcluster() with changed arguments, followed by an invocation of NUsetup() with a different 1st argument. To change only one's choice of y-Outcome variable and/or the Treatment/Exposure variable, a new NUsetup() invocation is all that is needed.

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2010) Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2015) NU_Confirm_Guidelines.pdf http://localcontrolstatistics.org

Obenchain RL. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

Rubin DB. (1980) Bias reduction using Mahalanobis metric matching. Biometrics 36: 293-298.

Tukey JW. (1977) Exploratory Data Analysis, New York: Addison-Wesley, Section 2C.

See Also

ltdagg, ivadj and lrcagg.

Examples

  
  # Running takes more than 7 seconds...
  data(pci15k)
  xvars   = c("stent", "height", "female", "diabetic", "acutemi", "ejfract", "ves1proc")
  hclobj  = NUcluster(pci15k, xvars)
  NU.env  = NUsetup(hclobj, pci15k, thin, surv6mo)
  surv050 = ltdagg( 50, NU.env)
  surv100 = ltdagg(100, NU.env)
  surv200 = ltdagg(200, NU.env)
  NUcompare(NU.env)
  

Specify KEY parameters used in NU.Learning to "design" analyses of Observational Data.

Description

Invoke NUsetup() to specify the name of the Hierarchical Clustering object output by NUcluster() and the name of the data.frame containing all desired X-covariates, the Treatment/Exposure variable and the Y-Outcome variable. It is ESSENTIAL to save the Environment output by NUsetup() as a named object within the user's .GlobalEnv space.

Usage

NUsetup(hclobj, dframe, trex, yvar)

Arguments

Value

The environment output by NUsetup() must be saved to the user's .GlobalEnv space. It's contents will be automatically updated by calls to other NU.Learning functions:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2010) Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

ltdagg, ivadj and lrcagg.

Examples

  
  # Running takes about 7 seconds...
  data(pci15k)
  xvars  = c("stent", "height", "female", "diabetic", "acutemi", "ejfract", "ves1proc")
  hclobj = NUcluster(pci15k, xvars)
  NUe    = NUsetup(hclobj, pci15k, thin, surv6mo)
  ls.str(NUe)
  

Confirm that Clustering in Covariate X-space yields an "adjusted" LTD/LRC effect-size Distribution

Description

For a given Number of Clusters, K, confirm() compares the observed distribution of LTDs or LRCs from relatively well-matched experimental units with the corresponding distribution from Purely Random Clusterings of experimental units. The larger are differences between the (blue) observed empirical CDF of effect-sizes and the (red) Purely Random CDF, the more potentially IMPORTANT are the "adjustments" resulting from focussing upon clustering (matching) of experimental units in X-space.

Usage

confirm(x, reps=100, seed=12345)

Arguments

Details

Making calls to confirm() for ltdagg() or lrcagg() objects resulting from different choices of K = Numbers of Clusters help the analyst decide which observed LTD or LRC effect-size distributions are (or are not) meaningfully different from Purely Random. When the X-covariates used in NUcluster() are truly "ignorable," then [i] all X-based clusters will be Purely Random, and [ii] both the number (K) and the sizes (N1, N2, ...,NK) of clusters formed will be meaningless and arbitrary. Thus the NU Strategy confirm() function simulates the empirical CDF for LTDs or LRCs resulting from purely random permutations of the Cluster ID numbers (1, 2, ...,K) assigned by ltdagg() or lrcagg(). Each permutation yields K artificial "clusters" of sizes N1, N2, ..., NK. Simulation results are accumulated for the total number of random permutations specified in the "reps=" argument of confirm(). Calls to print.confirm() and plot.confirm() provide information on comparisons of empirical CDFs for the Observed and Purely Random LTD/LRC distributions, including calculation of an observed two-sample Kolmogorov-Smirnov D-statistic using stats::ks.test. This is a non-standard use of ks.test() because the distributions being compared are DISCRETE; both contain many within-cluster TIED effect-size estimates. The p-value computed by ks.test() is not reported or saved because it is badly biased downwards due to TIED estimates. Researchers wishing to simulate a p-value for the observed KS D-statistic that is adjusted for TIES can invoke KSperm(confirm()).

Value

An output list object of class confirm:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2010) The Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

ltdagg and lrcagg.

Instrumental Variable LAO Fitting and Smoothing

Description

For a given number of patient clusters in baseline X-covariate space and a specified Y-outcome variable, smooth the distribution of Local Average Outcomes (LAOs) plotted versus Within-Cluster Propensity-like Scores: the Treatment Selection Fraction or the Relative Exposure Level.

Usage

  ivadj(x)

Arguments

Details

Multiple invocations of ivadj(ltdagg()) or ivadj(lrgagg()) using varying numbers of clusters, K, can be made. Each invocation of ivadj() displays a linear lm() fit and a smooth.spline() fit to the scatter of LAO estimates plotted versus their within-cluster propensity-like score estimates.

Value

An output list object of class ivadj:

Author(s)

Bob Obenchain <wizbob@att.net>

References

McClellan M, McNeil BJ, Newhouse JP. (1994) Does More Intensive Treatment of Myocardial Infarction in the Elderly Reduce Mortality?: Analysis Using Instrumental Variables. JAMA 272: 859-866.

Obenchain RL. (2010) Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

Rosenbaum PR, Rubin RB. (1983) The Central Role of the Propensity Score in Observational Studies for Causal Effects. Biometrika 70: 41-55.

See Also

ltdagg, lrcagg and NUcompare.

Examples

  
  # Running takes about 7 seconds...
  data(pci15k)
  xvars = c("stent", "height", "female", "diabetic", "acutemi", "ejfract", "ves1proc")
  hclobj = NUcluster(pci15k, xvars)
  NU.env = NUsetup(hclobj, pci15k, thin, surv6mo)
  surv050 = ltdagg(50, NU.env)
  iv050 = ivadj(surv050)
  iv050
  plot(iv050)
  

Calculate the observed Distribution of LRCs in NU.Learning

Description

For a given number, K, of Clusters of Experimental Units in baseline X-covariate space, lrcagg() calculates the observed distribution of "Local Rank Correlations" (LRCs) across Clusters ...where each LRC = cor(trex, Y, method = "spearman") within a Cluster, trex is a numeric measure of Exposure, and Y is a numeric measure of Outcome.

Usage

  lrcagg(K, envir)

Arguments

Details

Multiple calls to lrcagg(K) for varying numbers of clusters, K, are typically made after first invoking NUcluster() to hierarchically cluster patients in X-space and then invoking NUsetup() to specify a Y Outcome variable and a continuous, numerical treatment Exposure: trex. lrcagg() computes an observed LRC Distribution, updates information stored in its envir object, and outputs an object that is typically saved in the user's .GlobalEnv to allow subsequent use by print.lrcagg(), plot.lrcagg(), confirm() or KSperm(). Uninformative Clusters (those containing only 1 or 2 experimental units) contribute NA values to the LRCtabl$LRC and LRCdist$LRC objects within the lrcagg() output list.

Value

An output list of 12 objects, of class lrcagg:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2010) The Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2019) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

ivadj, ltdagg and NUcompare.

Examples

    data(radon)
    xvars = c("obesity", "over65", "cursmoke")
    hclobj = NUcluster(radon, xvars)
    e = NUsetup(hclobj, radon, lnradon, lcanmort)
    lrc050 = lrcagg(50, e)
    lrc050
    plot(lrc050, e)

Calculate the Observed Distribution of LTDs in NU.Learning

Description

For a given number, K, of Clusters of Experimental Units in baseline X-covariate space, ltdagg() calculates the observed distribution of "Local Treatment Differences" (LTDs) of the form LTD = (( mean(Y) for units receiving trtm==1 ) - ( mean(Y) for units receiving trtm==0 )).

Usage

  ltdagg(K, envir)

Arguments

Details

Multiple calls to ltdagg(K) for varying numbers of clusters, K, are typically made after first invoking NUcluster() to hierarchically cluster patients in X-space and then invoking NUsetup() to specify a Y Outcome variable and a two-level, numerical treatment variable: trtm. ltdagg() computes an observed LTD Distribution, updates information stored in its envir object, and outputs an object that is typically saved in the user's .GlobalEnv to allow subsequent use by print.ltdagg(), plot.ltdagg(), confirm() or KSperm(). Uninformative Clusters (those containing either only trtm==1 units or else only trtm==0 units) contribute NA values to the LTDtabl$LTD and LTDdist$LTD objects within the ltdagg() output list object.

Value

An output list of 12 objects, of class ltdagg:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2010) Local Control Approach using JMP. Chapter 7 of Analysis of Observational Health Care Data using SAS, Cary, NC:SAS Press, pages 151-192.

Obenchain RL. (2019) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

ivadj, lrcagg and NUcompare.

Examples

  
  # Running takes more than 7 seconds...
  data(pci15k)
  xvars = c("stent", "height", "female", "diabetic", "acutemi", "ejfract", "ves1proc")
  hclobj = NUcluster(pci15k, xvars)
  NUe = NUsetup(hclobj, pci15k, thin, surv6mo)
  surv050 = ltdagg(50, NUe)
  surv050
  plot(surv050, NUe)
  

Create a <<Most-Like-Me>> data.frame for a specified X-Confounder vector: xvec

Description

For a Given X-confounder Vector (xvec), sort all experimental units (eUnits) in an ltdagg() or lrcagg() output object into the strictly non-decreasing order of their distances from this X-Vector, which defines the TARGET eUnit: "Me". Plots of mlme() objects and displays of mlme.stats() are then used to Visualize and Summarize "Mini-" << LOCAL effect-size Distributions >> for different Numbers of "Nearest Neighbor" eUnits.

Usage

  mlme(envir, hcl, NUagg, xvec )

Arguments

Details

For example, in demo(radon), the eUnits are 2881 US "Counties", and the NUagg object is of type lrcagg() because radon exposure is a continuous variable. But, in demo(pci15k), the eUnits are 15487 "Patients," and the NUagg object is of type ltdagg() because treatment choice (thin) is Binary (0 = "No", 1 = "Yes").

Value

An output list object of class mlme:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. NU.Learning-vignette. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

plot.mlme,print.mlme,mlme.stats

Examples

  
  # Running takes about 7 seconds...
  data(pci15k)
  xvars = c("stent", "height", "female", "diabetic", "acutemi", "ejfract", "ves1proc")
  hclobj = NUcluster(pci15k, xvars)
  NU.env = NUsetup(hclobj, pci15k, thin, surv6mo)
  surv0500 = ltdagg(500, NU.env)
  xvec11870 = c( 0, 162, 1, 1, 0, 57, 1) 
  mlmeC5H = mlme(envir = NU.env, hcl = hclobj, NUagg = surv0500, xvec = xvec11870 ) 	
  plot(mlmeC5H) # using default "NN" and "breaks" settings...
  

Print Summary Statistics for One or More "Most-Like-Me" Histogram Pairs.

Description

Print Summary Statistics for Local effect-size (LTD or LRC) Distributions associated with given Numbers of "Nearest-Neighbors" in X-confounder Space.

Usage

mlme.stats(x, NN = 50, ...)

Arguments

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

See Also

plot.mlme,print.mlme,mlme

Six-month Survival, Cardiac cost and Baseline Covariate data for 15,487 PCI patients.

Description

Using observational data on 996 patients who received a Percutaneous Coronary Intervention (PCI) at Ohio Heart Health, Lindner Center, Christ Hospital, Cincinnati (Kereiakes et al, 2000), we generated this much larger dataset via "plasmode simulation."

Usage

data(pci15k)

Format

A data frame of 11 variables on 15,487 patients; no NAs.

patid

Patient ID number: 1 to 15487.

surv6mo

Binary PCI Survival variable: 1 => Survival for at least 6 months following PCI, 0 => Survival for less than 6 months.

cardcost

Cardiac related costs incurred within 6 months of patient's initial PCI; numeric value in 1998 dollars; costs were truncated by death for the 404 patients with surv6mo == 0.

thin

Numeric treatment selection indicator: thin = 0 implies usual PCI care alone; thin = 1 implies usual PCI care augmented by either planned or rescue treatment with a new blood thinning agent.

stent

Coronary stent deployment; numeric, with 1 meaning YES and 0 meaning NO.

height

Height in centimeters; numeric integer from 133 to 198.

female

Female gender; numeric, with 1 meaning YES and 0 meaning NO.

diabetic

Diabetes mellitus diagnosis; numeric, with 1 meaning YES and 0 meaning NO.

acutemi

Acute myocardial infarction within the previous 7 days; numeric, with 1 meaning YES and 0 meaning NO.

ejfract

Left ejection fraction; numeric value from 17 percent to 77 percent.

ves1proc

Number of vessels involved in the patient's initial PCI procedure; numeric integer from 0 to 5.

References

Kereiakes DJ, Obenchain RL, Barber BL, et al. Abciximab provides cost effective survival advantage in high volume interventional practice. Am Heart J 2000; 140: 603-610.

Gadbury GL, Xiang Q, Yang L, Barnes S, Page GP, Allison DB. Evaluating Statistical Methods Using Plasmode Data Sets in the Age of Massive Public Databases: An Illustration Using False Discovery Rates. PLOS Genetics 2008; 4: 1-8, e1000098 (Open Access).

Obenchain RL. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

Examples

    data(pci15k)
    str(pci15k)

Display an Instrumental Variable (LAO) plot with Linear and smooth.spline Fits

Description

For a given number of patient clusters, K, in baseline X-covariate space and a specified Y-outcome variable, display the distribution of Local Average Outcomes (LAOs) plotted versus Within-Cluster Propensity-like Scores: Treatment Selection Fractions or Relative Exposure Levels.

Usage

  ## S3 method for class 'ivadj'
plot(x, maxsiz = 0.15, ...)

Arguments

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

See Also

plot.ltdagg, plot.lrcagg

Display Visualizations of an Observed LRC Distribution in NU.Learning

Description

Display a Histogram, Box-Whisker Diagram and/or empirical Cumulative Distribution Function depicting the Observed Local Rank Correlation (LRC) Distribution across K Clusters.

Usage

  ## S3 method for class 'lrcagg'
plot(x, envir, show="all", breaks="Sturges", ...)

Arguments

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

See Also

plot.ltdagg

Display Visualizations of an Observed LTD Distribution in NU.Learning

Description

Display a Histogram, Box-Whisker Diagram and/or empirical Cumulative Distribution Function depicting the Observed Local Treatment Difference (LTD) Distribution across K Clusters.

Usage

  ## S3 method for class 'ltdagg'
plot(x, envir, show="all", breaks="Sturges", ...)

Arguments

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

See Also

plot.lrcagg

Display a Pair (or Pairs) of Histograms showing LOCAL effect-sizes for Patients "Most-Like-Me".

Description

Display Pair(s) of Histograms of Local effect-size (LTD or LRC) Distributions for a specified Number (or combinations of Numbers) of "Nearest-Neighbors in X-confounder Space.

Usage

  ## S3 method for class 'mlme'
plot(x, NN=50, breaks=50, ...)

Arguments

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

See Also

mlme.stats,print.mlme,mlme

Particulate Matter, Mortality and Other data for 2980 US Counties

Description

This data.frame combines 122 variables from the 5 sources referenced below. Several PM variables appear to be predictions from EPA “CMAQ” models rather than values from validated measuring instruments. NU.Learning concepts are illustrated in demo(pmdata) using Clustering of 2973 Counties and Parishes within the contiguous 48 US States and Washington, D.C.

Usage

data(pmdata)

Format

This data.frame contains 122 variables for 2,980 US counties. A total of 738 "NA"s imply that only about two tenths of one percent of these 363,560 values are missing.

fips

Federal Information Processing Standard code; 4 or 5 digits; 2980 unique values

C50

Cluster ID Number between 1 and 50. Total of 50 unique values

LRC50

Local (Spearman) Rank Correlation between Bvoc and AACRmort within Cluster

County

County or Parish name is a Factor variable (character code)

State

State name is a 2-Character Factor code; 49 unique levels

Deaths

CDC: Total number of Deaths in the County in 2016

Population

CDC: Total population of County in 2016

CRmort

CDC: Crude Rate of Circulatory-Respiratory Mortality for the County in 2016

CrudeL95

CDC: Lower 95% confidence limit for Crude Rate of CR Mortality

CrudeU95

CDC: Upper 95% confidence limit for Crude Rate of CR Mortality

CrudeSE

CDC: Standard Error for Crude Rate of CR Mortality

AACRmort

CDC: Age Adjusted Rate of Circulatory-Respiratory Mortality in 2016

AACRL95

CDC: Lower 95% confidence limit for Age Adjusted Rate of CR Mortality

AACRU95

CDC: Upper 95% confidence limit for Age Adjusted Rate of CR Mortality

AACR_SE

CDC: Standard Error for Age Adjusted Rate of CR Mortality

TotDeathPct

CDC: Total Death Percentage - Circulatory-Respiratory Mortality in 2016

lat

EPA: County Latitude used in EPA "CMAQ" model calculations

long

EPA: County Longitude used in EPA "CMAQ" model calculations

RHpct

EPA: Relative Humidity Percentage for 2016

SFCtmpC

EPA: Surface Temperature in Degrees Centigrade for 2016

NO2

EPA: Nitrogen Dioxide level (NO2.ppbV) for 2016

O3

EPA: Ozone level (O3.ppbV) for 2016

pmCL

EPA: Chlorine level in Particulate Matter (PM25_CL.ugm3) for 2016

pmEC

EPA: Ethylene Carbonate level in Particulate Matter (PM25_EC.ugm3) for 2016

pmNA

EPA: Sodium level in Particulate Matter (PM25_NA.ugm3) for 2016

pmMG

EPA: Magnesium level in Particulate Matter (PM25_MG.ugm3) for 2016

pmK

EPA: Potassium level in Particulate Matter (PM25_K.ugm3) for 2016

pmCA

EPA: Calcium level in Particulate Matter (PM25_CA.ugm3) for 2016

pmNH4

EPA: Ammonium level in Particulate Matter (PM25_NH4.ugm3) for 2016

pmNO3

EPA: Nitrate level in Particulate Matter (PM25_NO3.ugm3) for 2016

pmOC

EPA: Organic Compounds in pmTOT [fine particulate matter] (PM25_OC.ugm3) for 2016

pmOM

EPA: OM compounds in pmTOT (PM25_OM.ugm3) for 2016

pmOTHR

EPA: Other Compounds in pmTOT (PM25_OTHR.ugm3) for 2016

pmSO4

EPA: Sulfate Compounds in pmTOT (PM25_SO4.ugm3) for 2016

pmFE

EPA: Ferrous Compounds in pmTOT (PM25_FE.ugm3) for 2016

pmSI

EPA: Silicon Compounds in pmTOT (PM25_SI.ugm3) for 2016

pmTI

EPA: Titanium Compounds in pmTOT (PM25_TI.ugm3) for 2016

pmMN

EPA: Manganese Compounds in pmTOT (PM25_MN.ugm3) for 2016

pmAL

EPA: Aluminum Compounds in pmTOT (PM25_AL.ugm3) for 2016

pmUNSPCRS

EPA: UNSPCRS Compounds in pmTOT (PM25_UNSPCRS.ugm3) for 2016

pmPOA

EPA: Primary Organic Aerosols in pmTOT (PM25_POA.ugm3) for 2016

pmSOA

EPA: Secondary Organic Aerosols in pmTOT (PM25_SOA.ugm3) for 2016; pmSOA = Avoc+Bvoc

pmGLY

EPA: Glycemic Secondary Organic Aerosols in pmTOT (PM25_GLYSOA.ugm3) for 2016

pmOLGB

EPA: OLGB compounds in pmTOT (PM25_OLGB.ugm3) for 2016

pmISOP

EPA: ISOP compounds in pmTOT (PM25_ISOP.ugm3) for 2016

pmEPOX

EPA: EPOX compounds in pmTOT (PM25_EPOX.ugm3) for 2016

pmSQT

EPA: SQT compounds in pmTOT (PM25_SQT.ugm3) for 2016

pmMTN

EPA: MTN compounds in pmTOT (PM25_MTN.ugm3) for 2016

pmMT

EPA: MT compounds in pmTOT (PM25_MT.ugm3) for 2016

pmTOT

EPA: Total (fine) Particulate Matter (PM25_TOT.ugm3) for 2016

SFCtmpK

EPA: Surface Temperature in Degrees Kelvin for 2016

CardioRes

CDC: Cardio Respiratory Rate (rate I00J98.per100000.cdc) for 2016

POPcdc

CDC: County Population (population.cdc) for 2016

POP5yracs

CDC: 5yracs Population (population.people.5yracs) for 2016

PREMdeath

Premature Deaths per 100K Residents ...UWPHI for 2018

POFHealth

Poor or Fair Health rate (Poor.or.fair.health) ...UWPHI for 2018

PPHdays

Poor Physical Health days (Poor.physical.health.days) ...UWPHI for 2018

PMHdays

Poor Mental Health days (Poor.mental.health.days) ...UWPHI for 2018

LBW

Low Birth Weight rate (Low.birthweight) ...UWPHI for 2018

ASmoke

Adult Smoking Percentage (Adult.smoking) ...UWPHI for 2018

AObes

Adult Obesity Percentage (Adult.obesity) ...UWPHI for 2018

FEnv

Food Environment Index (Food.environment.index) ...UWPHI for 2018

PhysInAct

Physical Inactivity (Physical.inactivity) ...UWPHI for 2018

ExercOPS

Access to Exercise Opportunities ...UWPHI for 2018

ExsDrink

Excessive Drinking Rate (Excessive.drinking) ...UWPHI for 2018

AIDrivD

Alcohol Impaired Driving Deaths ...UWPHI for 2018

STInfect

Sexually Transmitted Infections ...UWPHI for 2018

TBirths

Teenage Births (Teen.births) ...UWPHI for 2018

Uninsur

Uninsured Residences (Uninsured) ...UWPHI for 2018

PCDocs

Primary Care Physicians (Primary.care.physicians) ...UWPHI for 2018

Dentists

Dentists (Dentists) ...UWPHI for 2018

PrevntHS

Preventable Hospital Stays (Preventable.hospital.stays) ...UWPHI for 2018

DiabMNT

Diabetes Monitoring (Diabetes.monitoring) ...UWPHI for 2018

MammoSC

Mammography Screening (Mammography.screening) ...UWPHI for 2018

SomCOL

Some College Education (Some.college) ...UWPHI for 2018

UnEMP

Unemployment Rate (Unemployment) ...UWPHI for 2018

ChildPOV

Children Living in Poverty (Children.in.poverty) ...UWPHI for 2018

IncomIEQ

Income Inequality (Income.inequality) ...UWPHI for 2018

ChildSPH

Children In Single-Parent Households ...UWPHI for 2018

SocASOC

Social Associations (Social.associations) ...UWPHI for 2018

VioCRM

Violent Crime Rate (Violent.crime) ...UWPHI for 2018

InjyDths

Injury Death Rate (Injury.deaths) ...UWPHI for 2018

AirPolpm

Air Pollution Particulate Matter ...UWPHI for 2018

DrnkWtVi

Drinking Water Violations (Drinking.water.violations) ...UWPHI for 2018

SevrHOUS

Severe Housing Problems (Severe.housing.problems) ...UWPHI for 2018

DrivATW

Driving Alone to Work (Driving.alone.to.work) ...UWPHI for 2018

LComutA

Long Commute - Driving Alone to Work ...UWPHI for 2018

PAAM

Premature Age Adjusted Mortality ...UWPHI for 2018

FrqPhysD

Frequent Physical Distress ...UWPHI for 2018

FrqMentD

Frequent Mental Distress ...UWPHI for 2018

DiabPrev

Diabetes Prevalence (Diabetes.prevalence) ...UWPHI for 2018

FoodInSec

Food Insecurity (Food.insecurity) ...UWPHI for 2018

LimAHFood

Limited Access to Healthy Foods ...UWPHI for 2018

DrugOdDM

Drug Overdose Deaths Model predictions ...UWPHI for 2018

InsufSlp

Insufficient Sleep (Insufficient.sleep) ...UWPHI for 2018

UnInsAds

Uninsured Adults (Uninsured.adults) ...UWPHI for 2018

UnInsCls

Uninsured Children (Uninsured.children) ...UWPHI for 2018

HCareCost

Health Care Costs (Health.care.costs) ...UWPHI for 2018

OthPrimCP

Other Primary Care Providers ...UWPHI for 2018

MHHIncome

Median Household Income ...UWPHI for 2018

ChildFRPL

Children Eligible for Free or Reduced-Price Lunch ...UWPHI for 2018

Population

County Population (Population) ...UWPHI for 2018

AGELess18

Residents below 18 Years of Age ...UWPHI for 2018

A65oOVR

Residents 65 or Older (X..65.and.older) ...UWPHI for 2018

NHispAfA

Non-Hispanic African-American Residents ...UWPHI for 2018

AmINalsN

American Indian or Alaskan Natives ...UWPHI for 2018

Asian

Asian Residents (X..Asian) ...UWPHI for 2018

NHawOPI

Native Hawaiian and Other Pacific Islanders ...UWPHI for 2018

Hispanic

Hispanic Residents (X..Hispanic) ...UWPHI for 2018

NHispWht

Non-Hispanic White Residents ...UWPHI for 2018

LoProEngl

Low Proficiency in English (not.proficient.in.English) ...UWPHI for 2018

Females

Female Residents ...UWPHI for 2018

Rural

Rural Residents ...UWPHI for 2018

pmOA

EPA: Organic Aerosols in pmTOT (PM25_OA.ugm3) for 2016

Avoc

EPA: Anthroprogenic [man-made] Volatile Organic Compounds in pmTOT for 2016

pmSEAspry

EPA: Sea Spray components in pmTOT (PM25_SOAAVOC.ugm3) for 2016

pmDUST

EPA: Dust components in pmTOT (PM25_DUST.ugm3) for 2016

pmNH4NO3

EPA: Ammonium Nitrate components in pmTOT (PM25_NH4NO3.ugm3) for 2016

pmSOOT

EPA: Soot components in pmTOT (PM25_SOOT.ugm3) for 2016

isop

EPA: SOA Isoprenes (PM25_SOAISOPRENE.ugm3) for 2016

terp

EPA: SOA Terpenes (PM25_SOATERPENE.ugm3) for 2016

Bvoc

EPA: Biogenic (natural) Volatile Organic Compounds for 2016; Bvoc = isop + terp

References

Obenchain RL. and Young SS. (2022), EPA Particulate Matter Data - Analyses using Local Control Strategy. (24 pages, 22 figures) https://doi.org/10.48550/arXiv.2209.05461

Pye, H., Ward-Caviness, C., Murphy, B., Appel, K., and Seltzer, K. (2021). Secondary organic aerosol association with cardiorespiratory disease mortality in the united states. Nature Communications, 12.7215 https://doi.org/10.1038/s41467-021-27484-1

Pye, H. [EPA] (2021), Data For Secondary Organic Aerosol and Cardiorespiratory Disease Mortality. https://doi.org/10.5281/zenodo.5713903

University of Wisconsin, Population Health Institute. https://uwphi.pophealth.wisc.edu [UWPHI] UWPHI@med.wisc.edu

Young SS. and Obenchain RL. (2022), "EPA particulate matter data...Analyses using Local Control Strategy" https://doi.org/10.5061/dryad.63xsj3v58

Examples

    data(pmdata)
    str(pmdata, list.len=122)

Print Summary Statistics on Local effect-size Estimates for Patients "Most-Like-Me".

Description

Display "Most-Like-Me" Summary Statistics for LOCAL effect-size (LTD or LRC) Distributions of "Nearest-Neighbors" in X-confounder Space.

Usage

  ## S3 method for class 'mlme'
print(x, ...)

Arguments

Value

NULL

Author(s)

Bob Obenchain <wizbob@att.net>

See Also

mlme.stats,plot.mlme,mlme

Radon exposure and lung cancer mortality data for 2,881 US counties in 46 States.

Description

Federal EPA and state government agencies have been reporting observational data at the US County level since about 1980. The data given here include 5 potential X-confounder variables of the relationship between lung cancer mortality and radon exposure; they were amassed and checked by Goran Krstic, Fraser Health Authority, Vancouver, BC, Canada.

Usage

data(radon)

Format

A data frame of 11 variables for 2881 US counties. One Missing Value; row 778 for Shannon County, SD, fips == 46113, has hhincome == NA.

fips

County FIPS code. Codes are 4 or 5 digit integers; 2881 unique values.

state

State Factor variable (2-character codes); 46 unique levels.

county

County or Parish Factor variable (character codes); 1703 unique levels.

lcanmort

Lung Cancer Mortality rate (deaths per 100,000 person-years), 1980-2004.

radon

County Radon Exposure level in picocuries per liter (pCi/L) for some unspecified period within 1986-1992; rounded to nearest single decimal place.

lnradon

Natural logarithm of County Radon Exposure level. Radon levels reported as 0.0 for 10 US counties are Windsorized here to ln(0.05), which is roughly -3.

obesity

Percentage of County Residents considered Obese (age adjusted), 2008.

over65

Percentage of County Residents of Age 65 and over, 2000 Census.

cursmoke

Percentage of County Residents who Currently Smoke, 1997-2003.

evrsmoke

Percentage of County Residents who Ever Smoked, 1997-2003.

hhincome

Average Median HouseHold Income in Thousands ($1,000), 1989-2004.

References

Krstic G, Obenchain RL. (2016) Radon dataset documentation and downloads. http://localcontrolstatistics.org

Obenchain RL. (2018) RADON_short.pdf http://localcontrolstatistics.org 40 PPT Slides and Commentary in Notes Pages format.

Examples

    data(radon)
    str(radon)

Create a data.frame for use in Prediction of a LTD/LRC effect-size Distribution

Description

reveal.data() forms a data.frame by sorting and appending the LTD or LRC exposure effect-size measures from ltdagg() or lrcagg() – as well as a Cluster membership-number variable – to a copy of the data.frame specified in NUsetup(). In the fourth and final REVEAL Phase of NU.Learning, a stretch-goal is to predict variation in LTD/LRC effect-size distributions using the known (baseline) X-covariate characteristics of experimental units. For example, the data.frame output by reveal.data() is suitable for input to party::ctree() as well as to a number of other "less Visual" prediction methods available in R.

Usage

reveal.data(x, clus.var="Clus", effe.var="eSiz")

Arguments

Value

The desired data.frame:

Author(s)

Bob Obenchain <wizbob@att.net>

References

Obenchain RL. (2023) NU.Learning_in_R.pdf http://localcontrolstatistics.org

See Also

ltdagg, lrcagg, and NUsetup.