Iatrogenic risk factors for cytomegalovirus retinitis

Iatrogenic Risks for CMV-R—Hodge et al 701

Iatrogenic risk factors for cytomegalovirusretinitis

William G. Hodge,*† MD, PhD, FRCSC;

Jean-François Boivin,† MD, ScD, FRCPC;

Stanley H. Shapiro,† PhD; Richard G. Lalonde,‡ MD, FRCPC;

Kirtida C. Shah,* MD; Bradley D. Murphy,* MD;

Michel A. Dionne,* MD; Aashish Goela,* MD

ABSTRACT • RÉSUMÉ

Background: To determine the iatrogenic risk factors for cytomegalovirus (CMV)retinitis in patients with the acquired immunodeficiency syndrome (AIDS).

Methods: The case–control study included 120 patients with CMV retinitis and 159patients without CMV retinitis, all of whom had AIDS and CD4 counts less than 50 cells/μL at the time of diagnosis of retinitis or at the defined corresponding datein the controls. Iatrogenic risk factors studied were corticosteroid use, treatmentwith chemotherapy, treatment with radiotherapy, and blood transfusions.

Results: Among the risk factors studied, only steroid use was predictive of CMVretinitis (odds ratio 6.41, 95% confidence interval 2.35–17.51). Based on this study,the use of steroids systemically elevated the risk of CMV retinitis.

Interpretation: This result may be useful to clinicians and health policy specialists in developing rational guidelines for screening, examinationfrequency, and targeted prophylaxis for patients with AIDS.

Contexte : Établir les facteurs de risque iatrogéniques pour la rétinite à CMV chezles patients atteints du syndrome d’immunodéficience acquise (SIDA).

Méthodes : Nous avons fait une étude cas-témoins, comprenant 120 patients quiavaient une rétinite à CMV et 159 patients sans la maladie.Tous étaient atteints duSIDA et avaient des comptes CD4 de moins de 50 cellules/μL au moment dudiagnostic de la rétinite pour les cas en question ou à la date correspondante decontrôle pour le cas témoins. Les facteurs de risque iatrogéniques ont comprisl’utilisation de corticostéroïdes, la chimiothérapie, la radiothérapie et les transfusionssanguines.

Résultats : Parmi les facteurs de risque étudiés, seuls les stéroïdes se sont avérésprédictifs de rétinite à CMV (rapport de cotes 6,41, intervalle de confiance 95 %2,35–17,51). Selon cette étude, l’utilisation systémique des stéroïdes augmente lerisque de rétinite à CMV.

From *the University of Ottawa Eye Institute, Ottawa, Ont., †theDepartment of Epidemiology and Biostatistics, McGill University,Montreal, Que., and ‡the Division of Infectious Diseases, Royal VictoriaHospital of McGill University Health Sciences Centre, Montreal, Que.

Originally received Oct. 27, 2004Accepted for publication Jul. 22, 2005

Correspondence to: Dr. William G. Hodge, University of Ottawa EyeInstitute, 501 Smyth Rd., Ottawa ON K1H 8L6; fax (613) 739-6139,[email protected]

This article has been peer-reviewed.

Can J Ophthalmol 2005;40:701–10

Cytomegalovirus (CMV) retinitis is one of themost devastating complications of the acquired

immunodeficiency syndrome (AIDS), and the mostcommon ocular opportunistic infection amongpatients with AIDS. It is a necrotizing, blinding formof retinitis that tends to occur late in the course ofAIDS. The reported lifetime cumulative incidence ofAIDS-related CMV retinitis ranges from 25% to40%.1 No other ocular opportunistic infectionamong patients with AIDS has a lifetime cumulativeincidence over 1% to 2%.2 Two large population-based studies demonstrated an increasing incidenceof this disorder in the late 1980s and the 1990s.3,4

Based on data from the era before highly active anti-retroviral therapy (HAART), the estimated meantime from development of AIDS to manifestation ofCMV retinitis is 9 to 18 months.5

With the advent of HAART,6 the incidence ofCMV retinitis is declining. Casado et al7 studied acohort of 172 patients with the human immunodefi-ciency virus (HIV)and a CD4 count less than 100 cells/μL at the time of protease inhibitor introduction. Thecumulative incidence of CMV retinitis was 5% at 1year and 6% at 2 years. Doan et al8 found that theannual risk of CMV retinitis in 1995 (without pro-tease inhibitors) was 6.1%, compared with 1.2% in1997 (with protease inhibitors); the annual risk ofrelapse after anti-CMV therapy decreased from 36%to 17%. Nevertheless, optimism should be temperedwith caution regarding this new treatment. At least20% of patients respond poorly or not at all toHAART,9,10 and the risk of CMV retinitis has likelynot changed for them. Furthermore, emerging resist-ance to a component of HAART is always possible,and the ophthalmic and public health communitiesshould be prepared for new CMV outbreaks at alltimes. Finally, in developing nations, AIDS and itsopportunistic infections are more devastating todaythan they have ever been. Hence, CMV retinitisremains one of the most important irreversibly blind-ing diseases in ophthalmology.

Several authors have investigated the treatment ofCMV retinitis,11–13 but fewer have examined riskfactors for this disease.14,15 Case–control studies thatcontribute to understanding these risk factors could

provide several clinical benefits. Increased under-standing may permit more sophisticated guidelinesfor screening, earlier detection, and a more logical algorithm for prophylactic therapy. One example of a drug for potential targeted prophylaxis is oral ganciclovir; it is seldom used however, because of itshigh cost and only moderate bioavailability.11,12

Valganciclovir, antiretroviral therapy, or experimentalimmunologic-based therapy may prove to play a rolein prophylaxis.13 However, the exact method ofoptimal targeted prophylaxis would need to be devel-oped in separate studies.

Iatrogenic risk factors for CMV infection of nonoc-ular tissue have been investigated in other immuno-suppressive settings, and these studies may provideclues for understanding the iatrogenic risk factors forCMV retinitis. One factor that has been implicated iscorticosteroid therapy. Nagler et al16 found a strongassociation between steroid use and development ofCMV pneumonia in bone-marrow transplant recipi-ents. Steroids have been found to be an importantrisk factor for CMV intestinal disease after intestinaltransplantation (in a historical cohort study by Manezet al17), for CMV disease in any organ in patientswith AIDS (Nelson et al18), and for CMV disease inrenal allograft recipients treated with cyclosporine(Morris et al19). Gorensek et al20 studied the riskfactors for CMV infection in heart transplant recipi-ents and found that the only significant predictor ofCMV infection was greater than average steroid use.

Immunosuppressive (nonsteroidal cytotoxic) therapyhas also been studied as a possible risk factor for sys-temic CMV disease. Immunosuppressive medicationuse and duration were the 2 most important non-serologic determinants of CMV infection after livertransplantation identified in a multicentre case–control study by Schiffer et al.21 Gorensek22 foundthe type and intensity of immunosuppression wereimportant risk factors after liver transplantation.More specifically, Kraat et al23 found triple immuno-suppressive therapy was an important risk factorafter renal transplantation. Kirklin et al24 analyzedthe cardiac transplant research database to look at riskfactors for CMV infection after heart transplantation.Among 1553 patients in the cohort, CMV disease

Iatrogenic Risks for CMV-R—Hodge et al

702 CAN J OPHTHALMOL—VOL. 40, NO. 6, 2005

Interprétation : Ce résultat peut aider les cliniciens et les spécialistes despolitiques de la santé dans la rédaction de lignes directrices rationnellespour le dépistage, la fréquence des examens et les objectifs de la pro-phylaxie à l’intention des patients atteints du SIDA.

developed in 230, and higher than average dosage of cytotoxic induction therapy agents was the mostimportant risk factor.

Some authors have observed an increase in CMVinfections in patients who have received multipleblood transfusions.25 Paya et al26 found transfusionsto be the most important risk factor for CMV infec-tion after liver transplantation, although Gorensek etal21 reported conflicting findings. The pathogenesisof this association is unclear, although it seems to bemore complicated than simple seroconversion.Possible mechanisms include transfer of an increasedviral load by the transfusions (CMV viremia); trans-fer of new, slightly modified strains of CMV thatcannot be determined with present serologic testing;or activation of preexisting CMV by factor(s) in thetransfusion. This last theory was championed bySloand et al25 in the discussion of their historicalcohort study looking at CMV infection of any organin patients with AIDS who had received blood trans-fusions.

Thus, there are several important gaps in ourknowledge with respect to the iatrogenic risk factorsfor CMV retinitis. Few studies have been performedto examine the risk factors for this important disease,and most of the studies investigating the etiology ofCMV retinitis have dealt with CMV disease ofnonophthalmic organs in non-AIDS, immunosup-pressive settings. The goal of this study was to deter-mine the iatrogenic risk factors for CMV retinitis ina large group of patients from 2 Canadian centres.

METHODS

A case–control study was conducted amongpatients with AIDS seen at the University of OttawaEye Institute and McGill University’s Department ofOphthalmology, Montreal, Que., between 1990 and1999. Cases were patients with a new diagnosis ofCMV retinitis who had a CD4 count below 50cells/μL at the time of diagnosis of retinitis. Controlswere patients without CMV retinitis but who had afirst CD4 count below 50 cells/μL. Cases and con-trols were not matched by year of inclusion in thestudy. The inclusion criteria were as follows: (1) HIVseropositivity confirmed by Western blot analysis, (2)a diagnosis of AIDS meeting the 1993 Centers forDisease Control definition of this disease,2 namely,HIV positive with at least 1 opportunistic infection,or HIV positive and absolute CD4 count less than200 cells/μL), (3) a CD4 count less than 50 cells/μL

at the time of inclusion in the study, and (4) age 18years or more at the index date (see index date defi-nition below).

Patients who met the eligibility criteria but werenot followed in the study base for at least 3 monthsbefore being eligible to be a case or control wereexcluded.

Sampling and study base

All eligible cases who received a diagnosis of CMVretinitis in the 2 study centres from 1990 to 1999were enrolled. Controls were sampled randomly fromthe study base at both institutions until the samplesize criterion was met, which was at least 198patients, based on an assumed odds ratio (OR) of 3.0for steroid use. The characteristics of the study base atthe 2 institutions were slightly different. In easternOntario, most of the patients with advanced HIVdisease are referred to the Immunodeficiency Serviceof the General Campus of the Ottawa Hospital. Mostof these patients are then sent to the Uveitis Serviceof the University of Ottawa Eye Institute for eyeexaminations whenever symptoms occur or occasion-ally for CMV surveillance eye examinations. Inwestern Quebec, one of the main sources of referralto the Uveitis Service of the Royal Victoria Hospitalis the Montreal Chest Hospital. Hence, the studybase for western Quebec consisted of patients withAIDS from the Montreal Chest Hospital who weresent to the Ophthalmology Department of the RoyalVictoria Hospital for ocular evaluation.

Index date

The date of diagnosis of retinitis represented theindex date for the cases. Risk factors for CMV retini-tis were then determined relative to this index date.We also needed to determine an index date for thecontrols. The most appropriate index date for thisgroup would have been the date corresponding to thesame duration of HIV infection as for the cases.However, this could not be determined accurately,since the exact time of HIV infection is typically dif-ficult to ascertain because initial infection with HIVmay be asymptomatic. We therefore decided torestrict our study population—cases and controls—to subjects with CD4 counts below 50 cells/μL. Ourassumption was that the duration of HIV infectionwould then be similar, on average, for both cases andcontrols. Specifically, the date of the first CD4 count


CAN J OPHTHALMOL—VOL. 40, NO. 6, 2005 703

recorded below 50 cells/μL represented the index datefor controls.

Exposure variables

Corticosteroid use

This variable was use of steroid, use of prednisone,and use of other steroids (all binary variables), andtotal amounts of these medications. Corticosteroidsare a well-known immunosuppressant.27 Based onprevious clinical studies,18–20 a risk period of 6 monthsbefore the index date was used in this study.

Chemotherapeutic agents

Other immunosuppressive chemotherapeutic agentsinvariably suppress leukocyte counts and were exam-ined as potential iatrogenic risk factors. Based onother studies that have looked at this exposure,23,24 arisk period of 3 months was chosen.

Radiotherapy

Some acute side effects of radiotherapy, includingskin and mucous membrane necrosis, typically occurin epithelial cells.28 However, there are also lateeffects, such as fistula formation, nonhealing ulcera-tion, and radiation retinopathy, which tend to occurin slowly metabolizing tissues. Hence, the risk periodchosen was 12 months.

Blood transfusion

The potential etiologic role of blood transfusion isnot clear.25 Explanations range from invoking aninfectious etiology to providing a cofactor for CMVinfection. In their clinical study, Sloand et al25 used 6months as the risk period when studying any CMVdisease in patients with AIDS. We also used 6 monthsas the risk period.

Confounders

The confounders chosen were variables associatedwith iatrogenic variables that also independentlyincrease or decrease the risk of retinitis. They wereCD4 count, anti-HIV medications, anti-CMV med-ications, anti-HSV (herpes simplex virus) medications,and hospital centre.

CD4 count

All patients had CD4 counts less than 50 cells/μL

at enrolment in the study. Based on our understand-ing of markers of disease severity, CD4 count beforeenrolment is the most important confounder in thisstudy. It is associated with illnesses that require iatro-genic treatments, such as those that we are studying.Furthermore, it is an extremely important predictorof CMV retinitis. The risk period for increased risk ofCMV retinitis in patients with a low CD4 count isknown to be short.5 This has been supported bystudies of CD4 counts and function that seem toindicate significant improvement in immune func-tion in the short term in patients receiving HAART.7

We therefore used the CD4 count measured in the 3months before the index date.

Anti-HIV medications

These medications are potentially associated withmany exposure variables because patients who needsteroid therapy, chemotherapy, radiotherapy, andblood transfusions are more likely to be taking moreanti-HIV medications and higher cumulative dosagesof these medications. Furthermore, patients receivinganti-HIV medication are at lower risk for CMVretinitis. We measured treatments received in the 2years before the index date.

Anti-CMV and anti-HSV medications

Anti-CMV medications used to treat non-ocularCMV infections were controlled for because thesemedications would be associated with predictors ofCMV retinitis and would protect against CMVretinitis itself. Furthermore, the anti-HSV medica-tion acyclovir has some anti-CMV activity29 andtherefore should be controlled for in a similar way.The etiologic period chosen was 2 years.

Hospital centre

Because there are immeasurable potential differ-ences between the hospital centres that may con-found the exposure–disease association, hospitalcentre was also controlled for. All data wereabstracted from the subjects’ medical records.

Statistical analysis

Univariate statistics were obtained to check forimplausible values as a final data editing step.Bivariate analyses were then conducted for all clinicalexposure variables using the unpaired t-test for con-



tinuous variables (or the Wilcoxon rank-sum test ifnon-normal) and the χ2 test for discrete variables.Logistic regression was used for multivariate analyses.A p value of 0.10 or less in the bivariate analyses wasused as a cutoff for inclusion in the multivariateanalyses. This cutoff was used so any variables thatmay be important in the outcome would not bemissed. Next, the assumption of linearity of the logitwas checked in a bivariate analysis for all exposureand confounding variables individually. This wasdone by graphing the ORs against each of the mid-points of multiple equal-interval variables. Forexample, CD4 count was divided into 4 approxi-mately equal intervals: 0–12, 13–24, 25–36, and37–49 cells/μL. Then the OR pertaining to casestatus corresponding to each interval was graphedagainst the midpoint of each interval (6, 18.5, 30.5and 43.0 cells/μL, respectively). From this plot, lin-earity of the logit was assessed. Next, a correlationmatrix was obtained for all exposure variables toassess multicollinearity. Any correlation coefficientgreater than 0.8 was recorded, and if a multi-collinearity problem arose in the multivariate model,an algorithm for omitting 1 variable was used. Thiscutoff was chosen so redundant variables would notbe used in the final model. This algorithm consistedof first keeping the variable that was a more biologi-cally plausible exposure, then keeping the variablewith fewer missing data, and finally keeping the vari-able with less measurement error as determined a priori.

For confounding, we determined a priori that themodel would include CD4 count, hospital site (city),at least 1 anti-HIV medication variable, at least 1anti-CMV medication variable, and at least 1 anti-HSV medication variable. Although these 5 cate-gories of confounding variables were chosen a priori,it was not clear for the 3 medication categories whatspecific representation of the variable to use (e.g., thespecific medication or the total dose of medication).Incorporating all possible representations of eachvariable would not be realistic because of precisionlimitations. Therefore, logistic regression was used asdescribed above to identify at least 1 variable for eachof the preselected drug treatment categories. Parallelmodels were used when different representations ofthe exposure variable were not independent. TheHosmer–Lemeshow goodness-of-fit test was usedwhen choices between parallel models were made.These confounding variables were then introducedinto the main exposure model.

Statistical techniques were not used to reduce thenumber of confounders in the model, but were usedto reduce exposure variables for the final model.Model reduction for the exposure variables was per-formed with the likelihood ratio test. If data weremissing, cases and controls were deleted in the multi-variate analysis. This is the simplest and most trans-parent way of dealing with missing data. However,when the model was reduced or expanded, these datawere then used again if the variable with the missingdata was eliminated from the model. The perform-ance of the final model against chance and its fit werethen determined.

The study protocol was approved by the require-ments of the Research Ethics Boards of both theOttawa Hospital and the Royal Victoria Hospital,Montreal Chest Hospital.

RESULTS

Demographic and descriptive results

There were 279 patients in the study, 148 fromMontreal and 131 from Ottawa. Of the 279, 120were cases (88 from Montreal) and 159 were controls



Table 1—Demographic and descriptive results

No. of subjects (%)

Cases (n = 120)

Controls(n = 159)

Hospital site Ottawa 32 (26.7) 99 (62.3)

Montreal 88 (73.3) 60 (37.7)

Gender

Male 109 (90.8) 136 (85.5)

Female 11 (9.2) 23 (14.5)

Caucasian 107 (89.2) 131 (82.4)

Index date, year

1990 6 (5.0) 5 (3.1)

1991 5 (4.2) 15 (9.4)

1992 15 (12.5) 13 (8.2)

1993 21 (17.5) 19 (11.9)

1994 23 (19.2) 33 (20.8)

1995 25 (20.8) 46 (28.9)

1996 20 (16.7) 23 (14.5)

1997 4 (3.3) 4 (2.5) 1998 1 (0.8) 1 (0.6)

(99 from Ottawa). Although cases and controls wereeligible for inclusion in the study between 1990 and1999, no cases were found in 1999. Table 1 summa-rizes the demographic and descriptive results for the2 groups. The average age at the index date was 38.6years for the cases and 39.6 years for the controls.Complete data were available for age, race, sex, treat-ment with radiotherapy, treatment with chemother-apy, medication use, CD4 count, and hospital site.For other variables, the completeness of the medicalrecords was as follows: blood transfusion 278(99.6%) and prednisone use 277 (99.3%).

Iatrogenic variables

The bivariate dichotomous data are summarized inTable 2. Based on the criteria summarized in the

methods section, all variables except treatment withchemotherapy were retained for the multivariate model.

The specific chemotherapeutic agents used in the 2groups are shown in Table 3. Table 4 presents the con-tinuous variables from the iatrogenic data. No vari-able was retained for the final clinical model.

Confounding by medication

It was decided a priori that at least 1 drug or drugamount from each of the anti-HIV, anti-CMV andanti-HSV drug categories would be used in the finalmultivariate model. The specific variables to beincluded were determined by logistic regression.

Table 5 summarizes the bivariate descriptive statis-tics and ORs for the dichotomous treatment vari-ables. The data for total number of anti-HIV med-ications and total duration of anti-HIV drug useamong cases and controls are shown in Table 6.



Table 2—Iatrogenic data for bivariate discrete variables

No. of cases (%)

No. of controls (%)

OR 95% CI pvalue

Steroids 23(19.2)

9(5.7)

3.95 1.78–8.75 0.0005

Prednisone 13(11.0)

7(4.4)

2.69 1.07–6.77 0.04

Chemotherapy 7(5.8)

7(4.4)

1.35 0.48–3.79 0.588

Radiotherapy 13(10.8)

4(2.5)

4.71 1.57–14.08 0.004

Blood transfusion 20(16.8)

12(7.6)

2.47 1.17–5.22 0.02

Note: OR, odds ratio; CI, confidence interval.

Table 3—Chemotherapeutic agents

No. of cases (%)

No. of controls (%)

Doxorubicin 4 (3.3) 2 (1.3)

Bleomycin 4 (3.3) 2 (1.3)

Vincristine 3 (2.5) 3 (1.9)

Vinblastine 2 (1.7) 2 (1.3)

Cyclophosphamide 0 (0.0) 2 (1.3)

Etoposide 0 (0.0) 1 (0.6)

Mitomycin C 0 (0.0) 1 (0.6)

5-Fluorouracil 0 (0.0) 1 (0.6)

Table 4—Continuous variables from iatrogenic model

p value Cases,mean ± SD

Controls,mean ± SD

Prednisonedosage, g

125.47± 495.86

34.75± 233.11

0.35

Radiotherapy dosage, Gy

2.20± 15.67

0.15± 1.92

0.77

No. blood transfusions

0.56± 1.42

0.95± 7.27

0.17

Note: SD, standard deviation.

In this bivariate analysis, cases took more drugs onaverage than controls, and the duration of medicationuse was longer among cases than among controls.

Whenever possible, the quantity of drug was alsorecorded for the risk period given in the methodssection. With some drugs, such as the proteaseinhibitors and foscarnet, too few observations wereavailable; therefore, the data are not presented forthese drugs. Those with reasonable precision are sum-marized in Table 7. Acyclovir was used in significantlygreater amounts among cases than among controls.

The final confounding model was then built

(Table 8). CD4 count, hospital centre, and treatmentvariables with a p value less than 0.10 were all initiallyforced into the model. The iatrogenic variablesretained from the univariate analysis and the con-founders shown in Table 8 were included. In the final



Table 5—Bivariate discrete statistics for dichotomous treatment variables

No. of cases (%)

No. of controls (%)

OR 95% CI p value

Anti-HSV 53 (44.2) 35 (22.0) 2.80 1.67–4.71 0.0001Anti-CMV 17 (14.2) 9 (5.7) 2.75 1.20–6.29 0.02 Ganciclovir 12 (10.0) 4 (2.5) 4.31 1.42–12.99 0.008Foscarnet 7 (5.8) 0 (0.0) — — 0.002 CMV test drug 3 (2.5) 6 (3.8) 0.65 0.18–2.44 0.55Anti-CMV combin-

ation therapy 6 (5.0) 0 (0.0) — — 0.004

Anti-HIV 93 (77.5) 106 (66.7) 1.72 1.01–2.95 0.05Zidovudine (AZT) 80 (66.7) 92 (57.9) 1.46 0.89–2.38 0.134 Didanosine (ddI) 47 (39.2) 60 (37.7) 1.06 0.65–1.73 0.81Zalcitabine (ddC) 30 (25.0) 37 (23.2) 1.10 0.63–1.91 0.74 Lamivudine (3TC) 27 (22.5) 19 (12.0) 2.14 1.31–4.04 0.02Stavudine (d4T) 5 (4.2) 1 (0.6) 6.87 1.04–12.32 0.04 Saquinavir 6 (5.0) 0 (0.0) — — 0.004Ritonavir 6 (5.0) 0 (0.0) — — 0.004 Indinavir 3 (2.5) 0 (0.0) — — 0.045

Note: OR, odds ratio; CI, confidence interval; HSV, herpes simplex virus; CMV, Cytomegalovirus;HIV, human immunodeficiency virus.

Table 6—Bivariate statistics for anti-HIV drugs and duration of drug use

Cases,mean ± SD

Controls,mean ± SD

pvalue

Number of anti-HIV medications

1.68± 1.41

1.32± 1.20

0.06

Duration of anti-HIV medication use, mo

12.82± 9.50

10.27± 9.50

0.02

Note: SD, standard deviation; HIV, human immunodeficiency virus; mo, months.

Table 7—Bivariate statistics for total amount of medication used

Cases, g mean ± SD

Controls, gmean ± SD

pvalue

Acyclovir 69.12± 126.20

37.28± 110.54

0.003

Ganciclovir 3.67 ± 18.17

0.67± 4.89

0.19

Zidovudine (AZT) 141.04± 149.50

125.71± 143.15

0.30

Didanosine (ddI) 45.86 ± 90.07

35.13± 64.68

0.61

Zalcitabine (ddC) 0.14± 0.38

0.11± 0.27

0.90

Lamivudine (3TC) 26.24 ± 60.86

12.54± 52.38

0.17

Note: SD, standard deviation.

multivariate model, only steroid use was a statisticallysignificant predictor of CMV retinitis (OR 6.41,95% confidence interval 2.35–17.51).

INTERPRETATION

From the iatrogenic model, several bivariate associ-ations were found to be predictive of CMV retinitis,including steroid use or blood transfusion within 6months of the diagnosis of CMV retinitis, andadministration of radiotherapy with a 12-month riskperiod. The use of chemotherapeutic agents was nota significant risk factor, nor was therapy with anyindividual agent. However, the power of the studywas not calculated for the assessment of this riskfactor, and the numbers were small, especially forindividual agents. Only steroid use was found to be asignificant predictor of CMV retinitis on multivariateanalysis. This finding is in keeping with other clinicalstudies of CMV risk factors in other immunosup-pressive settings.16–19

The biology and pharmacology of corticosteroidssupport their clinical role as immunosuppressants, aswell as their ability to play a part in the pathogenesisof CMV retinitis.27 They inhibit both the early(redness, edema, pain) and the late (leukocyte recruit-ment) components of the inflammatory response.One of their main mechanisms of action is to reducethe ability of neutrophils in vitro and in vivo toadhere to capillary and capillary-like tissue.Corticosteroids also modulate cytokine activity andother chemical mediators of inflammation.Corticosteroids seem to have a linear effect clinically.Small amounts equivalent to approximately 5 mg perday of prednisone are necessary to sustain life.Intermediate dosages, such as 20 mg of prednisone,have a moderate therapeutic effect and a moderate

side-effect profile. Larger dosages, such as 80 mg perday, have a profound anti-inflammatory andimmunosuppressive effect and a grave side-effectprofile.

This information will be useful for both theinternist and the ophthalmologist. Any HIV-positiveperson, especially those with CD4 counts less than 50cells/μL, who needs steroid therapy should be fol-lowed more carefully by an ophthalmologist. Suchpatients may be receiving steroids for chemotherapyprotocols or may need the drug because of restrictivelung disease from repeated bouts with Pneumocystispneumonia.

In any epidemiologic study, observed results maybe due to chance, bias, or confounding. We designedthis study to look for steroids as the main predictorvariable because that was our primary hypothesis atthe outset of the study. The power of the study wascalculated based on this assumption. Hence, we feelthat the study had adequate statistical power; the factthat our main predictor variable proved to be statisti-cally significant supports this.

The main biases of concern in this case–controlstudy were selection bias and misclassification bias.We should first point out that we chose acase–control design for 2 main reasons. First, itallowed us to examine multiple potential exposurevariables, which was one of the goals of the study.Second, with the advent of HAART in 1996, theincidence of CMV retinitis began declining to thepoint where it would not have been realistic to recruita cohort of unaffected patients that would provide astudy with adequate power. We attempted to mini-mize selection bias by adhering strictly to the studybase principle. In fact, for the Montreal patients, ourinitial hope was to use the patients at the ophthal-mology clinic of the Royal Victoria Hospital as our



Table 8—Final confounding model

OR SE 95% CI p value

Any ganciclovir use 3.11 1.95 0.91–10.61 0.07

Any acyclovir use 2.05 0.65 1.10–3.83 0.02

Number of anti-HIV drugs 1.43 0.18 1.13–1.82 0.03

CD4 count, cells/uL 0.96 0.01 0.94–0.98 0.0001

Hospital 5.19 1.62 2.82–9.58 0.0001

Note: OR, odds ratio; SE, standard error; CI, confidence interval.

study base. However, although cases were wellrecorded at the clinic, controls were not adequatelyretrievable from this study base. We thereforechanged our study base to patients referred from theMontreal Chest Hospital to the Royal VictoriaHospital to ensure that cases and controls would arisefrom the same population base, thereby respectingthe study base principle. We sought to prevent differ-ential misclassification bias by only including subjectswith at least 3 months of follow-up at the participat-ing clinics. Differential misclassification could haveoccurred if, for example, cases were followed moreclosely and more frequently before the index datethan controls; however, the minimum length offollow-up required in our study decreased the likeli-hood that this would occur. Nondifferential misclas-sification is always possible when medical records areused as the main source of information, but thiswould have biased our results toward the nullhypothesis, making the association found in thisstudy even more significant. Nevertheless, the smallnumber of patients receiving immunosuppressivecytotoxic drugs and the possible nondifferential mis-classification may be 2 reasons why no associationwas found between these drugs and the occurrence ofCMV retinitis.

We took great efforts to minimize the effect of con-founding. It was decided a priori to control for CD4count, as it is both the most important indicator ofdisease severity and an approximate indicator of timesince HIV infection, which was important for deter-mining the index date for controls. Hospital centrewas also controlled for because of the obvious poten-tial for confounding from one study base to the other.Regarding confounding by medication, we created amodel and decided a priori that it would contain atleast 1 variable for anti-HIV medication, 1 variablefor anti-CMV medication and at least 1 variable foranti-HSV medication (since anti-HSV medicationsmay also have some anti-CMV effects). This modelwas then used in the main exposure model to controlfor confounding. While it is impossible to ever beabsolutely certain that confounding has been com-pletely eliminated in any epidemiologic study, we feelconfident that we minimized its effect.

In conclusion, this study is one of the most com-prehensive studies to date examining the iatrogenicrisk factors for CMV retinitis. Steroid use representsan important risk factor for this disease in severelyimmunocompromised patients with AIDS.

The authors wish to acknowledge Dr. Jean Deschênes,Clinical Director, Department of Ophthalmology, RoyalVictoria Hospital, Montreal, Que., for helping with adminis-trative procedures for patient recruitment in Montreal.Supported by the University of Ottawa Young InvestigatorAward and the Medical Research Council of Canada.

REFERENCES

1. Holland GN, Pepose JS, Simons KB, et al. Ocular disease inthe acquired immune deficiency syndrome: clinicopathologi-cal correlations. In: Secchi AG, Fregona IA, eds. ModernTrends in Immunology and Immunopathology of the Eye.Milan, Italy: Saunders; 1989:132–55.

2. Holland G, Tufail A, Jordan MC. Cytomegalovirus diseases.In: Pepose JS, Holland GN, Wilhelmus KR, eds. OcularInfection & Immunity. New York, NY: Mosby; 1996:1088–129.

3. Montaner JS, Le T, Hogg R, et al. The changing spectrum ofAIDS index diseases in Canada. AIDS 1994;8:693–6.

4. Munoz A, Schrager LK, Bacellar H, et al. Trends in the inci-dence of outcomes defining acquired immunodeficiency syn-drome (AIDS) in the Multicenter AIDS Cohort Study:1985–1991. Am J Epidemiol 1993;137:423–38.

5. Pertel P, Hirschtick R, Phair J, Chmiel J, Poggensee L,Murphy R. Risk of developing cytomegalovirus retinitis inpersons infected with the human immunodeficiency virus. J Acquir Immune Defic Syndr 1992;5:1069–74.

6. McDonald CK, Kuritzkes DR. Human immunodeficiencyvirus type 1 protease inhibitors. Arch Intern Med 1997;157:951–9.

7. Casado JL, Arrizabalaga J, Montes M, et al. Incidence andrisk factors for developing cytomegalovirus retinitis in HIV-infected patients receiving protease inhibitor therapy. SpanishCMV-AIDS Study Group. AIDS 1999;13:1497–502.

8. Doan S, Cochereau I, Guvenisik N, Diraison MC,Mousalatti H, Hoang-Xuan T. Cytomegalovirus retinitis inHIV-infected patients with and without highly active anti-retroviral therapy. Am J Ophthalmol 1999;128:250–1.

9. Lewis JS, Terriff CM, Coulston DR, Garrison M. Proteaseinhibitors: a therapeutic breakthrough for the treatment ofpatients with HIV. Clin Ther 1997;19:187–214.

10. Le Moing V, Chene G, Carrieri MP, et al. Predictors of viro-logic rebound in HIV-1 infected patients initiating a proteaseinhibitor-containing regimen. AIDS 2002;16: 21–9.

11. Spector SA, McKinley GF, Lalezari JP, et al. Oral ganciclovirfor the prevention of cytomegalovirus disease in persons withAIDS. N Engl J Med 1996;334:1491–7.

12. Anderson RD, Griffy KG, Jung D, Dorr A, Hulse JD, SmithRB. Ganciclovir absolute bioavailability and steady-statepharmacokinetics after oral administration of two 3000-mg/ddosing regimens in human immunodeficiency virus- andcytomegalovirus-seropositive patients. Clin Ther1995;17:425–32.

13. Masur H, Whitcup SM, Cartwright C, Polis M, NussenblattR. Advances in the management of AIDS-related cyto-megalovirus retinitis. Ann Intern Med 1996;125:126–36.



14. Hodge WG, Boivin JF, Shapiro SH, et al. Laboratory-basedrisk factors for cytomegalovirus retinitis. Can J Ophthalmol2004;39:733–45.

15. Hodge WG, Boivin JF, Shapiro SH, et al. Clinical risk fac-tors for cytomegalovirus retinitis in patients with AIDS.Ophthalmology 2004;111:1326–33.

16. Nagler A, Elishoov H, Kapelushnik Y, Breuer R, Or R,Engelhard D. Cytomegalovirus pneumonia prior to engraft-ment following T-cell depleted bone marrow transplantation.Med Oncol 1994;11:127–32.

17. Manez R, Kusne S, Green M, et al. Incidence and risk factorsassociated with the development of cytomegalovirus diseaseafter intestinal transplantation. Transplantation 1995;59:1010–4.

18. Nelson MR, Erskine D, Hawkins DA, Gazzard BG.Treatment with corticosteroids—a risk factor for the develop-ment of clinical cytomegalovirus disease in AIDS. AIDS1993;7:375–8.

19. Morris DJ, Martin S, Dyer PA, Hunt L, Mallick NP, JohnsonRW. HLA mismatching and cytomegalovirus infection as riskfactors for transplant failure in cyclosporin-treated renal allo-graft recipients. J Med Virol 1993;41:324–7.

20. Gorensek MJ, Stewart RW, Keys TF, McHenry MC,Goormastic M. A multivariate analysis of the risk ofcytomegalovirus infection in heart transplant recipients. JInfect Dis 1988;157:515–22.

21. Schiffer V, Mentha G, Giostra E, Belli D, Le Coultre C,Rohner A. Cytomegalovirus in liver transplantation: inci-dence and groups at risk [in French]. Schweiz MedWochenschr 1994;124:631–6.

22. Gorensek MJ, Carey WD, Vogt D, Goormastic M. A multi-

variate analysis of risk factors for cytomegalovirus infection inliver-transplant recipients. Gastroenterology 1990;98:1326–32.

23. Kraat YJ, Christiaans MH, Nieman FH, van den Berg-Loonen PM, van Hooff JP, Bruggerman CA. Risk factors forcytomegalovirus infection and disease in renal transplantrecipients: HLA-DR7 and triple therapy. Transpl Int 1994;7:362–7.

24. Kirklin JK, Naftel DC, Levine TB, et al. Cytomegalovirusafter heart transplantation. Risk factors for infection anddeath: a multiinstitutional study. The Cardiac TransplantResearch Database Group. J Heart Lung Transplant 1994;13:394–404.

25. Sloand E, Kumar P, Klein HG, Merritt S, Sacher R.Transfusion of blood components to persons infected withhuman immunodeficiency virus type 1: relationship toopportunistic infection. Transfusion 1994;34:48–53.

26. Paya CV, Wiesner RH, Hermans PE, et al. Risk factors forcytomegalovirus and severe bacterial infections following livertransplantation: a prospective multivariate time-dependentanalysis. J Hepatol 1993;19:325–30.

27. Schimmer BP, Parker KP. Adrenocorticosteroids. In:Goodman LS, Gilman G, eds. The Pharmacologic Basis ofTherapeutics. 9th ed. New York, NY: McGraw Hill; 1996:1459–85.

28. Sausville EA, Longo DL. Principles of cancer therapy. In:Harrison TR, ed. Principles of Internal Medicine. 15th ed.New York, NY: McGraw-Hill; 2001:530–47.

29. Rachlis AR, Zarowny DP. Guidelines for antiretroviraltherapy for HIV infection. CMAJ 1998;158:496–505.

Key words: CMV retinitis, iatrogenic risk factors, corticosteroids



Documents

Iatrogenic risk factors for cytomegalovirus retinitis