The Relationship Between Delayed

or Incomplete Treatment and All-Cause

Mortality in Patients With Tuberculosis
Ariel Pablos-M\l=e'\ndez,MD, MPH; Timothy R. Sterling, MD; Thomas R. Frieden, MD, MPH

Objective.\p=m-\Toanalyze the factors associated with survival in patients with pul- people older than 5 years, particularly in
monary and extrapulmonary tuberculosis in New York City. Southeast Asia and sub-Saharan Africa.2
Design.\p=m-\Observationalstudy of a citywide cohort of tuberculosis cases. In the United States, a steady decline
in tuberculosis rates was reversed in the
Setting.\p=m-\NewYork City, April 1991, before the strengthening of its control pro- mid-1980s.3'4 In New York City, which
gram.
Patients.\p=m-\All229 newly diagnosed cases of tuberculosis documented by cul- reported nearly 15% of all US cases,5
the incidence of tuberculosis more than
ture in April 1991. Most patients (74%) were male, and the median age was 37 years doubled during the 1980s.6 The resurgence
(range, 1-89 years). In all, 89% belonged to minority groups. Human immunode- of tuberculosis has been attributed to hu¬
ficiency virus (HIV) infection was present in 50% and multidrug resistance in 7% of man immunodeficiency virus (HIV) in¬
the cases. Twenty-one patients (9%) were not treated. fection,7,8 increasing poverty,9 increasing
Main Outcome Measures.\p=m-\Follow-upinformation was collected through the immigration,10 dismantling ofcontrol pro¬
New York City tuberculosis registry; death from any cause was verified through the grams, and poor adherence to treatment,11
National Death Index. and the resurgence of tuberculosis has
Results.\p=m-\Cumulativeall-cause mortality by October 1994 was 44%; the median been accompanied by rising drug resis¬
survival for those who died was 6.3 months (range, 0 days to 3 years). The most tance.1214
important baseline predictors of mortality, adjusted for baseline clinical and demo-
graphic factors, were acquired immunodeficiency syndrome (AIDS) (91% vs 11% See also pp 1229 and 1259.
in HIV-seronegative patients; Cox relative risk [RR], 7.8; 95% confidence interval
[CI], 2.1-29.1), multidrug resistance (87% vs 39% in pansensitive cases; adjusted Several articles in the past 10 years
RR, 5.8; 95% CI, 2.3-14.5), and lack of treatment (81% vs 40%; adjusted RR, 3.1; have documented high case-fatality rates
95% CI, 1.0-9.7). Also, 11 of 13 HIV-infected patients who started treatment after among tuberculosis patients with HIV
a 1-month delay died. Among 173 patients surviving the recommended treatment infection15,16 and multidrug-resistant dis¬
period, those who completed therapy (66%) had a lower subsequent mortality (20% ease.1718 However, except for reports
vs 37%; RR, 0.5; 95% CI, 0.3-0.9). from selected hospitals,15,17 the determi¬
nants of patient survival during the cur¬
Conclusions.\p=m-\Mortalityfrom tuberculosis was high, even among patients rent tuberculosis resurgence have not
without multidrug resistance who were not known to be infected with HIV. Most
been studied. In April 1991, we ana¬
HIV-seropositive patients with delayed therapy died. Multidrug resistance predicted lyzed rates and risk factors for drug
higher mortality, and treatment completion was associated with improved subse- resistance among all patients reported
quent patient survival. with tuberculosis in New York City.12
JAMA. 1996;276:1223-1228 We now report the mortality of that
cohort after up to 3.5 years of follow-up.
IT IS ESTIMATED that one third of the
From the Division of General Medicine, College of PATIENTS AND METHODS
Physicians and Surgeons, Columbia University, New world's population is infected with My-
York, NY (Dr Pablos-M\l=e'\ndez);Bureau of Tuberculosis cobacterium tuberculosis.1 More than 7 The selection of patients and other
Control, New York City Department of Health (Drs million new cases occurred in 1990, re¬ methods used have been published.12
Pablos-M\l=e'\ndez,Sterling, and Frieden); Keesler Medi-
cal Center, Keesler Air Force Base, Miss (Dr Sterling); sulting in more than 2 million deaths, a Briefly, all patients in New York City
and Centers for Disease Control and Prevention, At- toll predicted to increase by the year 2000. ' who had a positive culture for M tuber¬
lanta, Ga (Dr Frieden). With 6% of all deaths worldwide attrib¬ culosis in the month of April 1991 (here¬
Reprints: Ariel Pablos-M\l=e'\ndez,MD, MPH, College of utable to tuberculosis, the disease is the after referred to as the index culture)
Physicians and Surgeons, Columbia University, 622 W
168th St, PH-9E-105, New York, NY 10032. leading infectious cause of death among were selected. All cultures were corrobo-

Downloaded From: http://jama.jamanetwork.com/ by a New York University User on 06/01/2015

Demographic and Clinical Characteristics and All-Cause Mortality Risk for 229 Patients During 3Vz Years lungs (including pleurisy, but not me-
Following a New Diagnosis of Tuberculosis in April 1991 in New York City* diastinal lymphadenopathy or a positive
Patients, Dead, No. RR (95% CI)
blood culture).19 Multidrug resistance
No. (%) (%) —I was defined as resistance to at least iso-
Characteristic (n=229) (n=101) Crude Cox-Adjusted t niazid and rifampin; if a patient's isolate
Age,y was susceptible to all 5 first-line agents
a35 150(65) 78 (52) 1.8(1.2-2.6) 1.8(1.0-3.3)
<35 79 (34) 23 (29) Reference
(isoniazid, rifampin, ethambutol hydro¬
Sex
chloride, pyrazinamide, and streptomy¬
Male 78
cin sulfate), it was considered pansen¬
170(74) (46) 1.2(0.8-1.7) 0.6(0.4-1.2)
Female 59 (76) 23 Reference
sitive (included are 9 cases not tested
(39) for pyrazinamide resistance). Poor com¬
Race/ethnicity
Asian 12(5) 1(8) 0.2(0.0-1.1) 0.4(0.1-3.8) pliance was defined as treatment de¬
African American 57 0.8(0.5-1.3) 1.1 (0.4-2.5) fault for 2 or more months. Patients who
130(57) (44)
were smear or culture positive after 5
Hispanic 62 (27) 30 (48) 0.9(0.6-1.5) 1.0(0.4-2.4)
White Reference
months of treatment were considered to
25(11) 13(52) have failed therapy. Patients were con¬
Country of birth sidered to have completed treatment if
United States 170(74) 83 (49) 1.6(1.0-2.4) 1.0(0.5-2.1)
59 (26) 18(30) Reference they received antituberculosis drugs
Homelessness regularly for 6 or more months (at least
Yes 58 (25) 29 (50) 1.2(0.9-1.6) 0.8(0.5-1.4) 9 months if HIV-seropositive and 12
No 171 (75) 72 (42) Reference months if multidrug resistant) and had
Alcoholism no microbiological or clinical evidence of
Yes 59 (26) 32 (54) 1.3(1.0-1.8) 1.3(0.7-2.3) tuberculosis during the last 3 months of
No 170(74) 69 (40) Reference treatment. A patient was considered
Injection drug use cured if mycobacteriology results were
Yes 39 1.3(0.7-2.5)
49 (22) (79) 2.3(1.8-2.9) negative at the end of treatment.
No 180(79) 62 (34) Reference
Mycobacteriologic results and treat¬
Extrapulmonary tuberculosis ment information were monitored
Yes 57 (25) 18(32) 0.6(0.4-1.0) 0.9(0.4-1.9) through the New York City tuberculo¬
No 172(75) 83 (48) Reference
sis registry. Mortality data were ascer¬
Cavitary diseased tained from the death certificate regis¬
Present 51 (23) 15(29) 0.6(0.4-1.0) 0.6(0.3-1.1)
Absent 172(77) 83 (48) Reference try of New York and verified through
the National Death Index, a federal reg¬
Acid-fastsmear§
Negative 38 (23) 17(45) 0.9(0.6-1.4) 1.1 (0.6-2.2) istry operated by the National Center
Positive 61 Reference
for Health Statistics.20 Survival was mea¬
129(77) (47)
HIV status
sured from the time the index culture
AIDS 59 7.9(2.7-23.1) 7.8(2.1-29.5) was plated to the time of death, regard¬
(26) 54(91)
HIV-seropositive 55 (24) 26 (47) 4.1 (1.4-12.3) 1.7 (0.4-6.4) less of its cause. Patients were followed
Unknownll 89 (39) 18(20) 1.7 (0.6-5.5) 1.1 (0.3-4.0) for up to 3.5 years (until October 1994);
Reference
survivors not seen after December 31,
HIV-seronegative 26(11) 3(11)
Drug resistance 1993, were censored on that date, the
Multidrug 15(7) 13(87) 2.2(1.7-2.9) 5.8(2.3-14.6) most recent date for which information
Other 31 (13) 16(52) 0.8(0.5-1.1) 1.1 (0.5-2.4) from the National Death Index was
None 72 (39) Reference available. For patients who died, the
183(80)
Treatment category median survival was 6.3 months (range,
Nontreated 21(9) 17(81) 2.0(1.5-2.6) 3.2(1.0-9.7) 0-40 months); for survivors, the median
Delay 30(14) 12(40) 1.0(0.6-1.6) 0.9 (0.3-2.5) follow-up time was 33.7 months (range,
Appropriate 178(77) 72 (40) Reference 32-42 months).
Statistical analysis included descrip¬
*RR indicates relative risk; CI, confidence interval; HIV, human immunodeficiency virus; and AIDS, acquired tive statistics of the study population
immunodeficiency syndrome. and bivariate analysis; Epi Info 621 and
fCox-proportional hazard models included all variables in the table,
ichest radiograph was not reported in 6 cases. SPSS/Windows22 were used. Categori¬
§Not tabulated are 62 patients with fewer than 2 smears reported. cal data were contrasted using the Man-
| Patients with no documented HIV serology and no AIDS-related conditions.
tel-Haenszel 2 test; when the expected
rated by the Centers for Disease Control Clinical and demographic data were value of a cell was less than 5, the Fisher
and Prevention mycobacteriology labo¬ obtained from multiple sources: medical exact test was used. To compare con¬
ratory, where DNA fingerprinting (re¬ records, submitting laboratories, the tinuous data, the t test was used for
striction fragment length polymorphism) New York City Bureau of Tuberculosis normally distributed variables; other¬
was performed on all available isolates. Control, the public shelter registry, and wise, the Wilcoxon 2-sample test was
Of the original 466 cases, 12 were ex¬ the New York City Health and Hospi¬ chosen. All tests of significance were
cluded because their index cultures were tals Corporation. Medical record reviews 2-tailed. Survival analysis included
determined to be contaminants by DNA were carried out by trained investiga¬ Kaplan-Meier curves and Cox propor¬
fingerprinting, clinical and laboratory re¬ tors or physicians. The study was ap¬ tional hazards modeling to adjust for
view, and epidemiologie analysis. Patients proved by the Institutional Review multiple variables23; the hazard ratio is
with relapse or patients who were un¬ Board of the New York City Depart¬ an expression of the mortality rate ra¬
dergoing treatment for more than 4 weeks ment of Health. tio. The proportional hazard assumption
(n=225) were also excluded, leaving a to¬ Extrapulmonary tuberculosis was de¬ was examined with time-dependent co-
tal of 229 new cases. fined as disease in organs other than the variate interactions. All variables in the

Downloaded From: http://jama.jamanetwork.com/ by a New York University User on 06/01/2015

 Table were entered in the multivariate
models because they are clinically rea¬
sonable and not highly collinear (all pair-
wise statistics were s0.42); backstep
elimination in multiple logistic analysis
yielded similar results with respect to
statistical significance and parameter es¬
timates in the full model. Pairwise in¬
teraction terms were not statistically
significant in the multivariate analysis.
RESULTS
Demographic characteristics for the
229 patients are shown in the Table.
Most patients (74%) were male, and the
median age was 37 years (range, 1-89
years). In all, 89% belonged to racial/
ethnic minorities. Clinical disease was
predominantly pulmonary (87%), al¬
though 25% had extrapulmonary tuber¬
culosis. Initial chest radiograph was re¬
ported as normal in 5 (5%) of 93 patients
who were HIV-seronegative or of un¬
known status and in 10 (10%) of 101
HIV-seropositive patients with pulmo¬
nary tuberculosis. Cavitary disease was
less common among HIV-seropositive
individuals (19% vs 34%; P<.02).
Fifty percent of the patients were
documented to be HIV-seropositive, and
half of these (59/114) had a diagnosis of
acquired immunodeficiency syndrome
(AIDS) by 1987 criteria other than tu¬
berculosis. Ofthe remaining 115 patients,
HIV test results were documented as
negative in only a quarter, but none had
evidence of HIV infection during follow-
up. Among the injection drug users, 94%
(46/49) were HlV-seropositive; the only
drug injection user with undocumented
HIV status was cured and alive at the
end of the follow-up period. A complete
analysis of baseline drug resistance has
been published.12 Human immunodefi¬
ciency virus infection was associated
with multidrug resistance.
Treatment and Follow-up
In 152 cases (66%), antituberculosis
therapy was started less than a week af¬
ter the date the index culture was plated.
Treated patients were started on a me¬
dian of 3 antituberculosis drugs. Only 10%
of patients received directly observed
therapy, which was usually prompted by
poor compliance. Twenty-one patients
(9%) were never treated: only 6 had posi¬
tive smears, 4 ofwhom refused treatment.
In 30 additional cases (13% of the 229),
treatment was delayed for 30 or more
days after the date the index culture was
plated. Two thirds (20) of these patients
had negative smears, and 11 had normal
chest radiographs.
Mycobacteriology follow-up specimens
were not routinely collected. Among 196
patients surviving 3 months after the
index culture, 6 were untreated and 83
Downloaded From: http://jama.jamanetwork.com/ by a New York University User on 06/01/2015
HIV-Seronegative (n=26) o Unknown
12
Follow-up Time,
Figure 1.—Kaplan-Meier survival curves for all patients with tuberculosis by human immunodeficiency virus
(HIV) status. Unknown indicates patients with no documented HIV serology and no acquired immunodefi¬
ciency syndrome (AIDS)-related conditions.
(42%) had no mycobacteriologic follow-
up. Of the 107 who had subsequent my¬
cobacteriologic results, 35 (18% of the
196 early survivors) had treatment fail¬
ure and in only 40 (20% of the 196 early
survivors) was cure bacteriologically
documented. Smears and cultures be¬
came consistently negative after a me¬
dian of 115 and 128 days, respectively, in
72 patients (37% of the 196). Of the 15
multidrug-resistant cases, 6 died in the
first month, 1 was not treated (and died),
4 others did not have mycobacteriologic
follow-up, and the available sputum re¬
sults of the remaining 4 patients re¬
mained positive for acid-fast bacilli.
Follow-up drug-susceptibility testing
was reported in 69 patients, of whom 9
(13%) developed new drug-resistance,
including multidrug resistance in 3; 7 of
these 9 cases were pansensitive at base¬
line. The possibility of reinfection was
not studied.
Seventy-five percent of patients (173/
229) survived the minimum period to be
able to complete therapy, including 72
(63%) of 114 HIV-seropositive and 101
(88%) of 115 HI V-seronegative/unknown
patients. Two thirds of these patients
completed treatment: 40 (56%) of 72
HIV-seropositive subjects and 74 (73%)
of 101 HI V-seronegative/unknown cases,
with median treatment duration of 516
and 345 days, respectively (P<.001).
Overall, 50% of the patients completed
antituberculosis treatment (114/229).
Just 1 of the 15 cases with multidrug
(n=89) +HIV-Seropositive (n=55) 'AIDS (n=59)
24 36 48
mo
resistance completed treatment (384
days under direct supervision). The only
documented tuberculosis relapse after
treatment completion occurred in a
homeless patient with AIDS who was
initially drug-susceptible M tuberculo¬
sis and who had poor compliance during
his 9-month treatment regimen.
Mortality
Survival was analyzed after up to 3.5
years of follow-up. Cumulative all-cause
mortality at 1 year was 28% (63/229); 17
(81%) of the 21 deaths among the 115
HIV-seronegative/unknown cases oc¬
curred in the first year, as opposed to
46 (57%) of the 80 deaths in the 114
HIV-seropositive patients (P<.001). By
October 1994, 101 patients (44%) had
died. The median survival for HIV-
seropositive patients (17 months; range,
0-40.6 months) was significantly shorter
than that for HIV-seronegative/un¬
known cases (33.4 months; range, 0-42
months) (P<.001).
Multidrug resistance, AIDS, and lack
of treatment were independent baseline
predictors of mortality, both in bivari-
ate analysis and in the multivariate
model (Table). Human immunodeficien¬
cy virus status was the most important
predictor of mortality (Figure 1). Of the
101 deaths, 80 (79%) occurred among
patients known to be infected with HIV.
Patients with AIDS had a cumulative
mortality of 91% (54/59), a median time
to death of 5.2 months (range, 0-39

MDR-TB(n=15)
Figure 2.—Kaplan-Meier survival curves for all patients with tuberculosis by drug resistance pattern.
TB indicates tuberculosis resistant to isoniazid and rifampin.
months), and a Cox-adjusted relative
risk (RR) of death, compared with HIV-
seronegative patients, of 7.8 (95% con¬
fidence interval [CI], 2.1-29.1). However,
3 of 26 patients with documented nega¬
tive HIV serology died (11%; 95% CI,
2.5%-30%), all men with pulmonary tu¬
berculosis. One ofthese patients refused
treatment and died within 3 months;
another had isolated isoniazid resistance
and died on the fourth day of treatment;
and the third completed therapy after
205 days and died 14 months later with¬
out evidence of tuberculosis.
Multidrug resistance was another im¬
portant independent predictor ofsurvival
(adjusted RR, 5.8; 95% CI, 2.3-14.6, com¬
pared with pansensitive cases), and its
effect was more evident in HIV-seroposi¬
tive patients (cumulative mortality of 92%
[12/13] vs 63% [54/85]; RR, 1.4; 95% CI,
1.2-1.8) than in HI V-seronegative/un¬
known patients (50% [1/2] vs 18% [18/98];
RR, 2.7; 95% CI, 0.6-11.6). The median
time to death for multidrug-resistant cases
was 4.2 months (range, 18 days to 3 years),
compared with 7.2 months (range, 0 days
to 3.3 years) for the other fatal cases
(P>.30). Resistance other than to both
isoniazid and rifampin did not have an in¬
dependent impact on survival (Figure 2).
As shown in the Table, early treat¬
ment was associated with improved sur¬
vival. Seventeen (81%) of the 21 un¬
treated patients died, 12 of them in the
first month (8 of whom had HIV infec¬
tion). Eleven of 13 HIV-infected patients
with delayed treatment died (median
Downloaded From: http://jama.jamanetwork.com/ by a New York University User on 06/01/2015
o Pansensitive (n=183) Other Pattern (n=31)
24 48
Follow-up Time, mo
MDR-
survival, 5.3 months; range, 1.3-34.7
months), compared with only 1 of 17
patients with delayed therapy who were
not known to be HIV-infected (RR, 14.4;
95% CI, 2.1-9.8). This interaction was
statistically nonsignificant, yet the ef¬
fect of suboptimal treatment is more
evident in HIV-seropositive individuals
(Figure 3).
To a lesser extent, older age was also
associated with higher mortality (adjusted
RR, 1.8; 95% CI, 1.0-3.3 for age >35 years
vs <35 years). Among HI V-seronegative/
unknown patients, increased mortality
was significant both for patients older
than 35 years (25% [17/69] vs 9% [4/46];
RR, 2.0; 95% CI, 1.0-7.8) and for alcohol¬
ics (33% [7/14] vs 15% [14/94]; RR, 2.2;
95% CI, 1.0-4.8); these interactions with
HIV status were not significant in the
multivariate model, although our sample
size was relatively small.
Among the 173 patients surviving the
minimum 6 to 12 months required to com¬
plete treatment, 45 (26%) died. In this
subgroup, HIV infection remained the
most important predictor of subsequent
mortality (53% [38/72] vs 7% [7/101]; RR,
7.6; 95% CI, 3.6-16.1). 4 of the 15
Only
patients with multidrug-resistant isolates
survived 12 months, and this factor did
not contribute to this analysis. Patients
who completed therapy (114/173,66%) had
a significantly lower subsequent
mortality (20% [23/114] vs 37% [22/59]; RR,
0.5; 95% CI, 0.3-0.9), with a statistically
nonsignificant trend persisting among
HIV-seropositive (47% [19/40] vs 59% [19/
32]; RR, 0.8; 95% CI, 0.5-1.2) and HIV-
seronegative/unknown patients
(5% [4/74] vs 11% [3/27]; RR, 0.5; 95% CI,
0.1-2.0). The effect of treatment comple¬
tion on subsequent mortality remained sig¬
nificant after adjusting for baseline de¬
mographic and clinical characteristics,
microbiological response, and poor ad¬
herence to treatment (RR, 0.3; 95% CI,
0.1-0.5) and after excluding the 27% of
deaths that occurred while patients were
receiving treatment (15% vs 31%; RR, 0.5;
95% CI, 0.3-0.9).
COMMENT
The results of this citywide investiga¬
tion illustrate the poor prognosis of tu¬
berculosis in the context of suboptimal
treatment. As reported previously,12 27%
of all patients with tuberculosis in New
York City in April 1991 died within 1
year of follow-up; after up to 3.5 years of
follow-up, cumulative mortality exceeded
44% (101/229). Although cause and effect
cannot be proven by our study, AIDS,
multidrug resistance, and delayed or in¬
complete therapy were important pre¬
dictors of mortality. The case-fatality rate
among HIV-seropositive patients with¬
out AIDS was also high but nonstatisti-
cally different from that rate in HIV-
seronegative individuals. The 18% (21/
117) case-fatality rate among patients not
known to be infected with HIV is re¬
markable, even if some of these patients
had early, unrecognized HIV infection;
only 1 of the 21 deaths in this subgroup
had multidrug-resistant M tuberculosis.
Among the 26 patients documented to be
HIV-seronegative, 3 (11%) died. A simi¬
lar case-fatality rate was reported in a
previous New York City cohort.15 The
mortality described in rural India before
the introduction of antituberculosis
therapy was 14.3% at 1.5 years24; the case-
fatality rate for US tuberculosis cases in
1984 was 7.8% to 9.7%.26
The indirect observational nature of
our study is both a limitation and a
strength. Some important measures (eg,
mycobacteriology follow-up or CD4+
T-lymphocyte counts) were not stan¬
dardized or systematically available, but
the variation in treatment implementa¬
tion could not have been realized under
any controlled protocol. The HIV sero¬
logical test results were not available in
most of the tuberculosis cases not oth¬
erwise thought to be infected; however,
our estimate of HIV infection (50%, 114/
229) coincides with those of other sur¬
veys in New York City.16·26 Also, only
culture-proven tuberculosis cases were
analyzed; in New York City, 5% to 17%
of tuberculosis cases are culture-nega¬
tive,5·7·27 and mortality rates may differ.
Survival follow-up was passive, and
actual cause of death was not deter-
 mined. Establishing tuberculosis as the
cause of death is difficult without an
autopsy and, consequently, death cer¬
tificates in New York have been found
to be unreliable.15,2i For this reason and
despite obvious limitations, the World
Health Organization defines mortality
rate by the number of tuberculosis cases
who die during treatment, regardless of
cause.19 The degree of immune deficiency
is the major determinant of mortality in
HIV-infected patients with tuberculo¬
sis,15·29 and the cause of death in treated
patients is generally AIDS, not tuber¬
culosis.30 On the other hand, among pa¬
tients without AIDS, tuberculosis itself
is held responsible for the less frequent
fatalities.31,32
Timely antituberculosis treatment is
equally effective in HIV-seropositive and
HIV-seronegative cases,29,30,33 although
relapses are more common among AIDS
patients.34 We observed that delayed an-
tituberculous therapy was followed by
disastrous consequences in immunocom¬
promised subjects, with 11 deaths oc¬
curring among 13 HIV-seropositive
cases whose treatment was delayed
more than 30 days. Treatment delays in
HIV-seronegative/unknown individuals,
on the other hand, usually did not lead
to death; indeed, before AIDS, 30% to
60% of tuberculosis patients survived 5
years without treatment.24 Untreated,
tuberculosis is a major contributor to
death in patients with AIDS.33·35 Ten of
the 13 untreated HIV-seropositive pa¬
tients in our series died within a month
of admission to the hospital. While the
severity of tuberculosis and associated
comorbidities were likely contributors
to this mortality, our results are con¬
sistent with clinical observations show¬
ing that therapeutic delays contribute
to early mortality in immunocompro¬
mised patients with tuberculosis.33·86
A causal connection between delayed
therapy and mortality in HIV-seroposi¬
tive patients is not the only possible
explanation for this association. It is pos¬
sible that the severity of tuberculosis at
time of diagnosis, by being related both
to mortality and therapeutic decisions,
could have confounded the association
between delayed therapy and mortal¬
ity. For this to have occurred, patients
with the most severe forms of tubercu¬
losis would need to have been those who
received delayed therapy; we believe
this is clinically implausible, since phy¬
sicians are generally more likely to start
empirical therapy earlier in sicker pa¬
tients. There is no validated index of
severity in tuberculosis, and this factor
is not included in the literature on tu¬
berculosis epidemiology and surveil¬
lance. We attempted to grade severity
by including extrapulmonary and cavi-
Downloaded From: http://jama.jamanetwork.com/ by a New York University User on 06/01/2015
-Timely Rx (n=88)
1.0 -r
> 0.5
12
Follow-up Time,
Figure 3.—Kaplan-Meier survival curves for human immunodeficiency virus-infected patients with tubercu¬
losis by treatment (Rx) timing. Delayed Rx indicates treatment was started more than 30 days after the in¬
dex culture was plated.
tary disease in the analysis, but these
variables were not independent predic¬
tors of mortality.
Additional explanations to the ob¬
served mortality among HIV-seroposi¬
tive patients with therapeutic delays in¬
clude the possibility that patients whose
treatment was delayed had more HIV-
associated comorbidities or were less
compliant with medication (even while
keeping clinic appointments). Although
treatment delays were frequently the
result of diagnostic oversight, some pa¬
tients may have been untreated because
they died too rapidly. Four of the 22
deaths among HIV-seropositive patients
with delayed or no treatment occurred
within 10 days of admission to the hos¬
pital. However, except for 1 patient dy¬
ing within 3 days of admission, the re¬
ported success of appropriate therapy
in people with AIDS30,35 suggests that
early antituberculosis treatment could
potentially have prolonged survival in
some or most of these patients.
Unfortunately, treatment delays were
not rare in this and other studies. In
some city hospitals during the 1980s,
20% of the patients with culture-proven
tuberculosis either died or were dis¬
charged without being diagnosed.36 In
the United States, approximately 6% of
cases of tuberculosis in the late 1980s
were diagnosed at death.5,37 A high in¬
dex of suspicion for tuberculosis needs
to be maintained and a lower therapeu¬
tic threshold should be used in HIV-
infected patients. Faster and more sen-
No Rx (n=13) -—
Delayed Rx (n=13)
24 36 48
mo
sitive laboratory methods to detect M
tuberculosis are needed to make a timely
diagnosis.
Whereas resistance to isoniazid alone
did not increase mortality, multidrug
resistance among HIV-seropositive pa¬
tients did, and 1 of the 2 HI V-seronega¬
tive/unknown patients with multidrug
resistant tuberculosis also died. Other
studies have indicated that resistance
to rifampin is associated with high fail¬
ure rates in tuberculosis treatment.38 In
a retrospective cohort of 171 refractory
cases of pulmonary tuberculosis resis¬
tant to isoniazid and rifampin, 63 pa¬
tients (37%) died despite expert man¬
agement.18 Fischi et al17 studied a cohort
of HIV-seropositive patients with tu¬
berculosis in Miami, Fla. The median
survival of 62 multidrug-resistant cases
was 2.1 months, compared with 14.6
months for 55 patients with single-drug-
resistant tuberculosis. It is possible that
multidrug-resistant tuberculosis, like
HIV infection, may be a marker for poor
adherence to treatment, riskier lives,
and other sociomedical characteristics
contributing to higher mortality. How¬
ever, the association between multidrug
resistance and mortality remained sig¬
nificant in our data after adjusting for
the social and behavioral indicators mea¬
sured, with impact mainly among HIV-
infected individuals.
The results of our study also highlight
the importance of completing antituber¬
culosis therapy as recommended.39 Among
patients surviving 6 to 12 months, those
 who completed therapy had half the sub¬
sequent mortality of those who did not.
The estimated benefit of completing
therapy did not vary after excluding pa¬
tients who died while receiving treatment
(ie, those in whom clinical deterioration
precluded treatment completion). Incom¬
plete therapy was primarily attributable
to noncompliance, which occurred in half
the cases. It is possible that treatment
default is just a marker for other con¬
founding variables such as poor under¬
lying health, noncompliance with zidovu¬
dine, or inferior health services. However,
many studies have documented the high
effectiveness of directly observed anti-
tuberculosis therapy even in patients with
AIDS,3*86 and the delinquent patients who
received such management in this study
were more likely to complete the antitu¬
berculosis regimen. Our data strengthen
References
1. Dolin PJ, Raviglione MC, Kochi A. Estimates of
future global tuberculosis morbidity and mortality.
MMWR Morb Mortal Wkly Rep. 1993;42:961-964.
2. Raviglione MC, Snider DE Jr, Kochi A. Global
epidemiology of tuberculosis: morbidity and mor-
tality of a worldwide epidemic. JAMA. 1995;273:
220-226.
3. Bloch AB, Reider HL, Kelly GD, Cauthen GM,
Hayden CH, Snider DE. The epidemiology of tu-
berculosis in the United States: implications for
diagnosis and treatment. Clin Chest Med. 1989;10:
297-313.
4. Jereb JA, Kelly GD, Dooley SW, Cauthen GM,
Snider DE. Tuberculosis morbidity in the United
States: final data, 1990. MMWR Morb Mortal Wkly
Rep. 1992;40(SS-3):23-27.
5. Centers for Disease Control. Tuberculosis in the
United States: 1987. Washington, DC: US Govern-
ment Printing Office; 1989. US Dept of Health and
Human Services publication CDC 89-8322.
6. New York City Dept of Health. Tuberculosis in
New York City, 1990: Information Survey. New
York, NY: New York City Dept of Health; 1991.
7. Mendelson MH, Adler J. Resurgence of tuber-
culosis: relationship to HIV and implications for
infection control. Mt Sinai J Med. 1990;57:221-224.
8. Barnes PF, Barrows SA. Tuberculosis in the
1990s. Ann Intern Med. 1993;119:400-410.
9. Drucker E, Alcabes P, Bosworth W, Sckell B.
Childhood tuberculosis in the Bronx, New York.
Lancet. 1994;343:1482-1485.
10. Cantwell MF, Snider DE Jr, Cauthen GM, On-
orato IM. Epidemiology of tuberculosis in the United
States, 1985 through 1992. JAMA. 1994;272:535-539.
11. Brudney K, Dobkin J. Resurgent tuberculosis
in New York City: human immunodeficiency virus,
homelessness, and the decline of tuberculosis con-
trol programs. Am Rev Respir Dis. 1991;144:745\x=req-\
749.
12. Frieden TR, Sterling T, Pablos-M\l=e'\ndezA, Kil-
burn JO, Cauthen GM, Dooley SW. The emergence
of drug-resistant tuberculosis in New York City.
N Engl J Med. 1993;328:521-526.
13. Bloch AB, Cauthen GM, Onorato IM, et al. Na-
tionwide survey of drug-resistant tuberculosis in
the United States. JAMA. 1994;271:665-671.
14. Neville K, Bromberg A, Bromberg R, Bonk S,
Hanna BA, Rom WN. The third epidemic\p=m-\multi-
drug-resistant tuberculosis. Chest. 1994;105:45-48.
15. Stoneburner R, Laroche E, Prevots R, et al. Sur-
vival in a cohort of human immunodeficiency virus-
infected tuberculosis patients in New York City: im-
plications for the expansion of the AIDS case
definition. Arch Intern Med. 1992;152:2033-2037.
Downloaded From: http://jama.jamanetwork.com/ by a New York University User on 06/01/2015
thewidely held belief that treatment
completion leads to improved subsequent
survival.
New York City has been at the epi¬
center of the HIV epidemic, the resur¬
gence of tuberculosis, and the emergence
of multidrug-resistant tuberculosis. Dur¬
ing 1990, only 50% of the patients with
tuberculosis in New York City completed
treatment.6 In Harlem, 158 (89%) of 178
patients with tuberculosis discharged in
1988 did not return to the clime and failed
to complete therapy.11 The premature
death of hundreds of young men and wom¬
en in the early 1990s is an unfortunate
lesson to the public health system and
should not be ignored in other parts of
the world. These developments prompted
federal, state, and local efforts to rees¬
tablish effective tuberculosis control pro¬
grams. Directly observed therapy has be-
16. Braun MM, Cot\l=e'\TR, Rabkin CS. Trends in
death with tuberculosis during the AIDS era.
JAMA. 1993;269:2865-2868.
17. Fischl MA, Daikos GL, Uttamchandani RB, et
al. Clinical presentation and outcome of patients
with HIV infection and tuberculosis caused by mul-
tiple-drug-resistant bacilli. Ann Intern Med. 1992;
117:184-190.
18. Goble M, Iseman MD, Madsen LA, Waite D,
Ackerson L, Horsburgh CR Jr. Treatment of 171
patients with pulmonary tuberculosis resistant to
isoniazid and rifampin. N Engl J Med. 1993;328:
527-532.
19. WHO Tuberculosis Programme. Framework for
Effective Tuberculosis Control. Geneva, Switzer-
land: World Health Organization; 1994. World Health
Organization publication WHO/TB/94.179.
20. Stampfer MJ, Willet WC, Speizer FE. Test of
the National Death Index. Am J Epidemiol. 1984;
119:837-839.
21. Dean AG, Dean JA, Coulombier D, et al. Epi
Info, Version 6. Atlanta, Ga: Centers for Disease
Control and Prevention; 1994.
22. Norusis MJ. SPSSfor Windows: Advanced Sta-
tistics, Release 6.0. Chicago, Ill: SPSS Inc; 1993.
23. Cox DR. Regression methods and life tables.
J R Stat Soc B. 1972;34:187-220.
24. Narain R, Gothi GD, Nair SS, et al. Tubercu-
losis in a rural population of South India: a five year
epidemiologic study. Bull World Health Organ. 1974;
51:473-488.
25. Centers for Disease Control. Tuberculosis Sta-
tistics: States and Cities, 1985. Washington, DC:
US Government Printing Office; 1986. US Dept of
Health and Human Services publication CDC 87\x=req-\
8249.
26. Centers for Disease Control. National HIV Se-
roprevalence Surveys: Summary of Results
(Through 1989). Washington, DC: US Government
Printing Office; 1990. US Dept of Health and Hu-
man Services publication HIV/CID/9-90/006.
27. Bureau of Tuberculosis Control. Information
Summary 1994. New York, NY: New York City
Dept of Health; 1995.
28. Washko RM, O'Doherty J, Frieden TR. Tuber-
culosis mortality, New York City, 1992. In: Pro-
ceedings of the 43rd Annual Epidemic Intelligence
Service Conference; April 18-22,1994; Atlanta, Ga.
29. Ackah AN, Coulibaly D, Digbeu H, et al. Re-
sponse to treatment, mortality, and CD4 lympho-
cyte counts in HIV-infected persons with tuber-
culosis in Abidjan, C\l=o^\ted'Ivoire. Lancet. 1995;345:
607-610.
30. Chaisson RE, Schecter GF, Theuer CP, Ruth-
come the standard of care.35·39"41 In New
York City, treatment completion rates
increased to approximately 90% by 1994.42
There was a 21% decline in the number
of new tuberculosis cases and a 60% de¬
crease in new multidrug-resistant isolates
from 1992 to 1994.27 Telzak et al43 recently
documented a therapeutic response in 24
of 25 HI V-seronegative patients with mul¬
tidrug-resistant tuberculosis, with only 1
fatal case (undiagnosed) after 91 weeks of
follow-up. Health care professionals and
policymakers should continue intensify¬
ing efforts to eliminate tuberculosis and
its accompanying mortality.
The authors thank Bruce Levin, PhD, and Wei
Yann Tsai, PhD, for their assistance with survival
analysis, Charles Knirsch, MD, and Richard De-
fendini, MD, for their valuable comments on this
work, and Steven Shea for his guidance and review
of the manuscript.
erford GW, Echenberg DF, Hopewell PC. Tuber-
culosis in patients with the acquired immunodefi-
ciency syndrome: clinical features, response to
therapy, and survival. Am Rev Respir Dis. 1987;
136:570-574.
31. Shirai T, Sato A, Chida K, et al. A study of
causes of death among patients with active pulmo-
nary tuberculosis from the standpoint of host fac-
tors. Kekkaku. 1990;65:397-405.
32. Nunn P, Brindle R, Carpenter L, et al. Cohort
study of HIV infection in patients with tuberculosis
in Nairobi, Kenya: analysis of early (6-month) mor-
tality. Am Rev Respir Dis. 1992;146:849-854.
33. Small PM, Schecter GF, Goodman PC, Sande
MA, Chaisson RE, Hopewell PC. Treatment of tu-
berculosis in patients with advanced human immu-
nodeficiency virus infection. N Engl J Med. 1991;
324:289-294.
34. Perri\l=e"\nsJH, St Louis ME, Mukadi YB, et al.
Pulmonary tuberculosis in HIV-infected patients
in Zaire: a controlled trial of treatment for either 6
or 12 months. N Engl J Med. 1995;332:779-784.
35. Alwood K, Keruly J, Moore-Rice K, Stanton
DL, Chaulk CP, Chaisson RE. Effectiveness of su-
pervised, intermittent therapy for tuberculosis in
HIV-infected patients. AIDS. 1994;8:1103-1108.
36. Mathur P, Sacks L, Auten G, Sall R, Levy C,
Gordin F. Delayed diagnosis of pulmonary tuber-
culosis in city hospitals. Arch Intern Med. 1994;
154:306-310.
37. Rieder HL, Kelly GD, Bloch AB, Cauthen GM,
Snider DE Jr. Tuberculosis diagnosed at death in
the United States. Chest. 1991;100:678-681.
38. Mitchison DA, Nunn AJ. Influence of initial
drug-resistance on the response to short course
chemotherapy of pulmonary tuberculosis. Am Rev
Respir Dis. 1986;133:423-430.
39. American Thoracic Society. Treatment of tu-
berculosis and tuberculosis infection in adults and
children. Am J Respir Crit Care Med. 1994;149:
1359-1374.
40. Chaulk CP, Moore-Rice K, Rizzo R, Chaisson
RE. Eleven years of community-based directly ob-
served therapy for tuberculosis. JAMA. 1995;274:
945-951.
41. Iseman MD, Cohn DL, Sbarbaro JA. Directly
observed treatment of tuberculosis: we can't afford
not to try it. N Engl J Med. 1993;328:576-578.
42. Frieden TR, Fujiwara PI, Washko RM, Ham-
burg MA. Tuberculosis in New York City\p=m-\turning
the tide. N Engl J Med. 1995;333:229-233.
43. Telzak EE, Sepkowitz K, Alpert P, et al. Mul-
tidrug-resistant tuberculosis in patients without
HIV infection. N Engl J Med. 1995;333:907-911.